Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

Acta Mechanica Sinica

https://doi.org/10.1007/s10409-020-00973-0

RESEARCH PAPER

Double diffusive convection in the finger regime for different Prandtl


and Schmidt numbers
Yantao Yang1

Received: 18 April 2020 / Revised: 1 June 2020 / Accepted: 10 June 2020


© The Chinese Society of Theoretical and Applied Mechanics and Springer-Verlag GmbH Germany, part of Springer Nature 2020

Abstract
In this work fingering double diffusive convection, i.e. the buoyancy-driven flow with fluid density being affected by two
different scalar components, is investigated numerically with special efforts on the influences of the physical properties of
two scalar components. We show that different scalar properties can affect the global transport behaviors. The concentration
flux exhibits different exponents in their power-law scalings for different combinations of scalar components. The scaling
exponents of heat flux, however, depend mainly on the ratio of the diffusivities of two scalars. If one uses the local parameters
of the finger layer in the bulk, the behaviors are very similar to those found in the fully periodic simulations. The horizontal
width of the fingers is consistent with the wavelength of the fast growing mode. For one case we observe evidences of the
thermohaline staircase, namely, the typical width of the flow structures changes significantly in different layers within the
flow domain.

Keywords Double diffusive convection · Convection turbulence · Turbulent mixing

1 Introduction A key point which distinguishes DDC from the normal


Rayleigh–Bénard convection is that the two scalar com-
Double diffusive convection (DDC) refers to the buoyancy- ponents have very different diffusivities. The difference in
driven convection flow where the fluid density depends on diffusivity allows various new instability mechanisms to
two scalar components with very different molecular dif- develop even when the overall density is stably stratified. In
fusivities. DDC occurs in many natural environments and the upper part of the (sub-)tropic oceans, both temperature
engineering applications, including astrophysics [1,2], geo- and salinity decrease with depth and fingering instability can
science [3,4], and process engineering [5]. The most relevant occur [15]. Here the flow is driven by the unstable salinity
terrestrial environment is the ocean since the density of sea- gradient and stabilized by the temperature gradient, which
water is determined by both temperature and salinity, and is in the fingering regime. In the upper part of many polar
their diffusivities differ by two orders of magnitude. Indeed, regions, both temperature and salinity increase with the depth
DDC has been widely observed in oceans [6–9]. It is esti- and the DDC happens in the diffusive regime.
mated that over 40% of the Earth ocean favors the occurrence It is very challenging to study DDC by experiments or
of DDC [10]. DDC can greatly enhance the diapycnal mix- numerical simulations. Experiments for DDC require con-
ing [7], and may even have influences on ocean climate [11]. trolling two scalar fields, which can be difficult. Depending
Therefore, it has drawn a lot of research interests in the on the exact experimental setup and the aim of the study,
past [12,13]. A comprehensive review can be found in the different combinations of scalar components were used in
recent book of Radko [14]. the past, including the heat-salt system [16], the heat-sugar
system [17], the salt–sugar system [18], and the recent heat-
copper-ion system [19]. The ratio between the diffusivities of
two scalars ranges from 3 for the salt-sugar system to over 200
B Yantao Yang
for the heat-copper-ion system. In numerical simulations, the
yantao.yang@pku.edu.cn
main challenge is to deal with the very small diffusivity of
1 SKLTCS and Department of Mechanics and Engineering concentration field such as salt, which requires very fine reso-
Science, BIC-ESAT, College of Engineering, and Institute of lution and huge amount of computing resources. A common
Ocean Research, Peking University, Beijing 100871, China

123
Y. Yang

practice in conducting three-dimensional direct numerical layer height H , the free-fall velocity of concentration field

simulations of DDC is to use a larger diffusivity for the con- gβC ΔC H , and the two scalar differences ΔT and ΔC . Then
centration field [20,21]. four control parameters are of relevance, saying the Prandtl
Then a natural question which needs to be answered is and the Schmidt numbers
how the different combinations of scalar diffusivities affect
the dynamics of fingering DDC. Some theoretical works has ν ν
Pr = , Sc = , (5)
been done to discuss the influences of the different diffusiv- κT κC
ities [22], but to our knowledge a systematic study on the
effects of the scalar diffusivities has not been reported yet. and two Rayleigh numbers
The present work presents preliminary results towards this
direction. Specifically, we conduct new simulations for dif- gβT ΔT H 3 gβC ΔC H 3
RaT = , RaC = . (6)
ferent combinations of scalar components at very high scalar νκT νκC
gradients. In combination with our previous datasets [23,24],
we focus on how the transport and the finger structures behave Two other commonly used parameters, which are the com-
for different scalar diffusivities. binations of the above four, are the Lewis number Le =
The paper is organized as follows. In Sect. 2 we briefly Sc/Pr and the density ratio Λ = (βT ΔT )/(βC ΔC ) =
describe the governing equations and numerical methods. In (Sc RaT )/(Pr RaC ). Thus, Le is the ratio between two
Sect. 3 we present our main results. Finally conclusions are scalar diffusivities, and Λ is the ratio of the density anomaly
given in Sect. 4. induced by the temperature difference to that by the con-
centration difference. Larger Λ indicates that the relative
strength of the stabilizing temperature difference is stronger.
2 Governing equations and numerical To numerically solve (1–4), we utilize our in-house
methods finite-difference code with a fraction-time-step method. An
efficient double-grid method is included to deal with the
We consider a layer of fluid bounded from top and bottom by concentration field with large Sc [25]. The code has been
two parallel plates. The two plates are separated by a height extensively applied to buoyancy-driven turbulence and wall-
H and perpendicular to the direction of gravity. We employ turbulence. In the present study, the two plates have constant
the Oberbeck-Buossinesq approximation in our governing temperature and concentration. Both no-slip and free-slip
equations, namely the density depends linearly on two scalar boundary conditions are investigated for the two plates. In
fields as ρ = ρ0 [1 − βT (T − T0 ) + βC (C − C0 )]. Here, ρ the horizontal directions periodic conditions are applied.
is density, T is temperature, C is concentration, βT is the The horizontal size of the domain is set to be at least 10
thermal expansion coefficient, and βC is the solutal contrac- times larger than the finger width, which can be estimated
tion coefficient, respectively. The subscript “0” denotes the as the half of the wavelength for the fastest growing mode
value at the reference point. Moreover, the approximation d = π Le1/4 /[RaC (Λ−1)]1/4 [21,26]. Initially both scalars
assumes that the two scalars only generate buoyancy force have linear distributions between two plates, and small ran-
in the momentum equation. Then, the governing equations dom perturbations are added to trigger the flow.
read Three different combinations of (Pr , Sc) are consid-
ered, namely (1, 100), (7, 70), and (7, 700), respectively. The
∂t u i + u j ∂ j u i = −∂i p + ν∂ 2j u i + gδi3 (βT T − βC C), (1) Rayleigh number of concentration RaC gradually increases
∂t T + u j ∂ j T = κT ∂ 2j T , (2) from 109 to 1012 . The Rayleigh number of temperature
increases accordingly so that the density ratio is fixed at
∂t C + u j ∂ j C = κC ∂ 2j C, (3) Λ = 1.6, which lies in the range usually observed in the
∂i u i = 0. (4) oceans. For (Pr , Sc) = (7, 700), i.e. the typical values in
ocean, we adopt the data with both no-slip and free-slip plates
Here, u i with i = 1, 2, 3 are the three components of veloc- from our previous works [23,24]. For the other two combi-
ity, p is pressure, ν is kinematic viscosity, g is gravitational nations of (Pr , Sc), only the free-slip boundary condition is
acceleration, δi j is the Kronecker Delta, κT and κC are the considered. Thus, four sets of cases are simulated with either
diffusivities of two scalars, respectively. different (Pr , Sc) or velocity boundary conditions at plates.
To drive the flow, we maintain a constant temperature The resolution is chosen such that the mesh size of the base
difference ΔT and a constant concentration difference ΔC grid is smaller than the viscous scale for the momentum field,
between the two plates. The top plate has higher tempera- and that of the refined grid is smaller than the Batchelor scale
ture and concentration so that the flow is in the fingering for the concentration fields. The numerical settings and key
regime. The flow quantities are nondimensionalized by the responses of the systems are summarized in Appendix.

123
Double diffusive convection in the finger regime for different Prandtl and Schmidt numbers

3 Results a
(1,100), F (7,700), F
3 (7, 70), F (7,700), N
10
3.1 Scalar transport

We first look at the global responses of the system. In the


mixing process of DDC, key responses include the fluxes of
2
two scalars and the flow velocity, which are measured by the 10
non-dimensional Nusselt and Reynolds numbers as
   10
9
10
10
10
11
10
12
N u T = −u 3 T V ,t κT ΔT H −1 , (7)
  
b
N u C = −u 3 CV ,t κC ΔC H −1 , (8)
1
Re = u r ms H /v. (9) 10

Here ·V ,t stands for the average over the entire domain and
time. u r ms is the root-mean-square (rms) value of velocity 10
0

magnitude. Note that we only include the convective flux in (1,100), F (7,700), F
the definitions of Nusselt numbers. In our simulations the (7, 70), F (7,700), N

Nusselt numbers are calculated by two different methods,


10 9 10 10 10 11 10 12
i.e. by their definitions or by using the exact global balance
relations [23,27]
c 10 3
 

T ≡ κT [∂i T ]2 = κT (ΔT )2 H −2 (N u T + 1), (10)


 V 2

C ≡ κC [∂i C]2 = −κC (ΔC )2 H −2 (N u C + 1).


10
(11)
V

The final value is taken as the average of the results from two 10 1 (1,100), F (7,700), F
methods. (7, 70), F (7,700), N
In Fig. 1 we show the dependences of the two Nusselt
9 10 11 12
numbers and the Reynolds numbers on the concentration 10 10 10 10
Rayleigh number. Our previous study shows that the con-
centration flux N u C is mainly determined by RaC for fixed Fig. 1 Global responses of the system versus RaC . a Concentration flux
(Pr , Sc) = (7, 700) [23]. The present results reveal that N u C , b Heat flux N u T , and c Reynolds number Re. Different colors
for each combination of (Pr , Sc), a scaling law still exists and symbols mark the different combinations of (Pr , Sc) and boundary
conditions, as indicated in the legend, where F stands for free-slip and
between N u C and RaC as N u C ∼ RaCα . However, the value N for no-slip, respectively. The dashed lines indicate the corresponding
of the scaling exponent α changes for different (Pr , Sc). By power-law scalings with the exponents given next to them
linear fitting we obtain α = 0.26 for (Pr , Sc) = (1, 100),
0.22 for (7, 70), and 0.31 for (7, 700) with both free-slip
and no-slip boundary conditions, respectively. The heat flux In many existing studies with triply periodic domains,
N u T and the Reynolds number Re also show clear power- such as those in Refs. [20,22], the only key control param-
law-scaling dependences on RaC , see Fig. 1b, c. The scaling eter is the density ratio Λ of the finger layers. However, it
laws of N u T divide into two distinct groups with different should be pointed out that the density ratio in fully periodic
exponents. The three sets of cases with Le = 100 exhibit domain is different from the global density ratio in the cur-
very similar exponents which is about 0.27. For the set rent simulations. In periodic domain, the finger layer is the
with Le = 10 the exponent is considerably smaller, with only flow structure due to the periodic boundary condition.
a value of 0.22. The exponent for the power-law scalings of In our bounded domain, two boundary layers develop adja-
Re shows relatively weak variations among different sets of cent to the two plates at top and bottom, and the finger layer
cases, which are all close to 0.4. Interestingly, Fig. 1 sug- only occupies the bulk region between the two boundary lay-
gests that for fixed (Pr , Sc) = (7, 700), velocity boundary ers. The global density ratio is measured across the domain,
condition, i.e. either free- or no-slip ones, does not affect i.e. the combined effects of the bulk finger layer and two
the scaling exponents for all three quantities. Compared to boundary layers. This can be clearly seen in the mean scalar
the no-slip boundary, Free-slip boundary does allow larger profiles shown in Fig. 2. For cases with RaC = 1010 , the
velocity and therefore stronger heat and concentration fluxes. mean temperature profiles exhibit very minor deviation from

123
Y. Yang

a 1 b 1 a
(1,100), F (7,700), F
10 2
(7, 70), F (7,700), N
0.8 0.8

0.6 0.6 1
10

0.4 0.4

9 10 11 12
10 10 10 10
0.2 0.2

b 0.8
0 0 (1,100), F (7,700), F
0 0.5 1 0 0.5 1 (7, 70), F (7,700), N
0.7

Fig. 2 Mean profiles of temperature and concentration for two cases 0.6
with RaC = 1010 and free-slip boundary condition. a (Pr , Sc) =
(7, 70), and b (Pr , Sc) = (7, 700), respectively 0.5

the linear profile over the whole domain height. This devia- 0.4

tion becomes more distinct for even higher RaC . The mean 1 2
10 10
concentration profiles, though, clearly reveal the two bound-
ary layers with very high gradient next to the top and bottom
c
plates, and the finger layer in the bulk with a linear profile. (1,100), F (7,700), F
The gradient of the mean concentration in the middle finger (7, 70), F (7,700), N
3
10
layer decreases as Le increases from 10 in Fig. 2a to 100 in
2b.
To solely measure the properties of the finger layers in
the bulk, we define the relevant non-dimensional parameters
by using the measured values in the middle of the domain, 2
10
saying 0.4 < z/H < 0.6. The mean gradients of T and C in
f f
this region, i.e. Tz and C z , are determined by the linear fit-
ting. Then the density ratio of the finger layer is calculated as 10
1
10
2
f f
Λ f = (βT Tz )/(βC C z ). The concentration Nusselt number
f f
of the finger layer is N u C = −u 3 C/(κC C z ). Hereafter 
Fig. 3 a Density ratio of the bulk finger layer Λ f versus the global
denotes the average over the horizontal plane and time. The concentration Rayleigh number RaC . b, c Density flux ratio γ f and
density flux ratio is defined as γ f = βT u 3 T /βC u 3 C, f
the local concentration Nusselt number N u C of the bulk finger layers
which is the ratio of the density anomaly transported by the versus Λ f

temperature flux to that by the concentration flux and com-


monly used for finger layers. Note that all quantities with the
superscript “ f ” are calculated only by the data in the region We then look at the fluxes of the finger layers in the bulk. In
0.4 < z < 0.6. Fig. 3b the finger-layer density flux ratio γ f is plotted against
The properties of the finger layers in the bulk are shown Λ f . All data points divide into two different groups with
in Fig. 3. As RaC increases, the density ratio of the fin- different Le. For the set with (Pr , Sc) = (7, 70), γ f first
ger layer Λ f decreases, indicating a relatively weaker mean decreases then increases as Λ f increases, which is a general
temperature gradient in the bulk, see Fig. 3a. For the global behavior for finger layers in fully periodic simulations [20,
parameters considered here, the measured density ratios of 22]. For all other three sets with Le = 100, the data points
the finger layers are quite large compared to the typical value follow a single trend despite the different Pr and velocity
discovered in the Oceans, saying 1 < Λ f < 2. Λ f also boundary conditions. It is very interesting to notice that this
shows a strong dependence on both Pr and Sc. Generally, dependence of γ f on Λ f is very similar to those found in
cases with bigger Pr and Sc results in a finger layer with the fully periodic domain, e.g. the results with (Pr , Sc) =
larger Λ f . (7, 700) [22]. For Le = 100 the minimal of γ f happens

123
Double diffusive convection in the finger regime for different Prandtl and Schmidt numbers

around Λ f = 4, which is smaller than all the values found 3.2 Finger structures
here.
As shown in Fig. 3c, the local concentration Nusselt num- In this subsection we discuss in details about the finger struc-
f
ber N u C decreases as Λ f for fixed (Pr , Sc), which is also tures in the bulk. Figure 4 demonstrates those structures of
similar to the finger layers in the periodic domain. Interest- three different cases by the contour of vertical velocity u 3 on
ingly, different behaviors can be observed at large and small a vertical plane. Vertically oriented fingers fulfill the whole
Λ f for (Pr , Sc) = (7, 700) with free- or no-slip bound- bulk. Slender fingers have small widths and relatively larger
f
ary conditions. At small Λ f , N u C has very similar values height. The ratio between the finger height to width changes
for different boundary conditions. While large discrepancy significantly for different combinations of Pr and Sc.
f
in the value of N u C exists at large Λ. Figure 3a indicates To measure the width of the fingers, we use the u 3 field
that large local density ratio Λ f are for the cases with small on the horizontal plane at the mid height z/H = 0.5. An
RaC . Small RaC corresponds to small ΔC or H . When ΔC example of such field is shown in Fig. 5 for (Pr , Sc) =
or H is small, the fingering structures in the bulk cannot fully (7, 70) and RaC = 1012 , which clearly shows the horizontal
develop before they reach the boundary layer. When RaC is sections of the finger structures. From these fields the finger
large enough, fingers can fully develop within the bulk and width is extracted by using the horizontal one-dimensional
the exact boundary condition at the plates does not matter autocorrelation function
anymore. And the local density ratio solely determines the
u 3 (x + δr)u 3 (x)
concentration flux. Φh (δr ) = , (12)
u 3 (x)u 3 (x)

where x is the coordinates on the horizontal plane, and


δr = |δr| with δr being the vector connecting two points
between which the correlation coefficient is computed. In
Fig. 6 we plot the autocorrelation function calculated for the
case shown in Fig. 5. The finger width d f is measured by
the δr of the first minimal in the curve, i.e. the most prob-
able distance between two fingers moving in the opposite
directions.
The finger width d f is calculated for all cases and pre-
sented in Fig. 7. The fingers become thinner as RaC increases.
Moreover, the dependence of d f on the global concentration
Rayleigh number RaC is consistent with that predicted by the
−1/4
wavelenght of the fastest growing mode, i.e d f ∼ RaC .
f
For the same RaC , d shows a strong dependence on the
ratio of diffusivities Le = Sc/Pr , instead of the exact val-
ues of Pr and Sc. Specifically, the three sets of cases with

Fig. 4 Finger structures shown by the vertical velocity fields on the


mid plane of the x-direction. Panels from let to right are three cases
with (Pr , Sc) = (1, 100) (left), (7, 70) (midlle), and (7, 700) (right),
respectively. All of them have RaC = 1012 and the free-slip bound-
ary conditions. The plots are shown with the original aspect ratios in Fig. 5 Contour of the vertical velocity u 3 on the horizontal plane at the
simulations mid height for the case with (Pr , Sc) = (7, 70) and RaC = 1012

123
Y. Yang

1.0

0.8

0.6

0.4

0.2

-0.2

0 0.01 0.02 0.03 0.04

Fig. 6 Autocorrelation function Φh for the case shown in Fig. 5, namely


(Pr , Sc) = (7, 70) and RaC = 1012
Fig. 8 Mean profiles for the case with (Pr , Sc) = (7, 70) and RaC =
1012 . a Scalar profiles, and b The rms profiles for two velocity compo-
-1
10 (1,100), F (7,700), F nents. The shaded area marks the middle finger layer
(7, 70), F (7,700), N

(1,100), F (7, 70), F (7,700), F (7,700), N

0.012

10 -2 0.010

0.008
9 10 11 12
10 10 10 10
0.006

Fig. 7 Finger width d f versus the global concentration Rayleigh num- 0 0.2 0.4 0.6 0.8 1
−1/4
ber RaC . The dashed line marks the power-law scaling of d f ∼ RaC
Fig. 9 The finger width d f at different heights z for the four cases at
RaC = 1012

Le = 100 have very similar finger widths, which is about


twice of the width for Le = 10. two very thin boundary layers adjacent to the two plates, there
are two regions with much smaller scalar gradients compared
3.3 Possible staircase state to the linear segment of the finger layer. The lower one locates
in the region 0 < z < 0.2, while the upper one in 0.8 < z <
One of the most striking phenomena induced by fingering 1.0, respectively. In Fig. 8b we also show the profiles of the
DDC is the thermohaline staircase, which is widely observed rms values for the horizontal velocity magnitude u rhms and
in the sub-tropic oceans [7] and been realized in both exper- the vertical velocity u rz ms . The rms values of the the velocity
iments and fully periodic numerical simulations [18,20]. In components are significantly larger in the two regions above
the staircase state, the flow domain consists of a stack of well- and below the middle finger layer than those within the finger
mixed convection layers separated by finger layers with high layer. Therefore, in these two regions the mean scalar profiles
scalar gradients. While for all the cases simulated in the cur- have smaller gradients and both the horizontal and vertical
rent study, one finger layer occupies the whole bulk, as shown velocities are stronger, indicating a local mixing at higher
in Fig. 4. However, one particular case shows some signs of level.
the emerging of staircase, saying with (Pr , Sc) = (7, 70) Difference between these two better mixed regions and
and RaC = 1012 . the finger layer in between can already be visually observed
To demonstrate this, we first look at the mean profiles for in the instantaneous flow field on the vertical plane, see the
this case, as given in Fig. 8. The mean profiles of temperature middle panel of Fig. 4. The horizontal length scale is clearly
and concentration are given in Fig. 8a. In the middle of the different at different heights of the domain. To quantitatively
bulk, i.e. roughly 0.2 < z < 0.8, the two scalars exhibit demonstrate the varying of the horizontal length scale, we
perfect linear profiles, which corresponds to the finger layer. calculate the finger width d f at different height z by using
However, above and below this middle finger layer and the the correlation functions (12). Indeed, the width d f is much

123
Double diffusive convection in the finger regime for different Prandtl and Schmidt numbers

Table 1 Summary of numerical parameters and global responses

d f (×10−2 )
f
Type Pr Sc RaC Γ Nx , Nz mx , mz tstat N uT N uC Re Λf N uC γ f

Free-slip 1 100 109 1.2 288,384 3,2 200 1.370 131.6 39.04 16.0 134.5 0.484 6.07
1010 0.6 288,576 3,3 200 1.717 253.6 98.26 9.58 264.0 0.479 3.52
1011 0.4 432,864 3,3 180 2.347 470.6 237.2 6.35 506.1 0.483 2.07
1012 0.2 1024,2048 2,3 76 3.341 798.7 541.1 4.89 739.8 0.467 1.19
Free-slip 7 70 109 2.0 384,384 4,2 400 4.578 90.80 26.26 4.39 97.17 0.647 3.86
1010 1.0 576,576 2,2 400 7.061 159.8 64.59 3.68 163.5 0.612 2.17
1011 0.6 768,1152 2,2 200 11.06 266.4 153.1 3.26 234.8 0.596 1.19
1012 0.3 768,2304 2,2 120 16.88 414.6 355.5 2.90 327.7 0.606 0.671
Free-slip 7 700 109 0.7 256,384 4,4 320 1.515 135.5 5.065 105 140.6 0.666 6.66
1010 0.5 384,768 4,4 200 1.919 290.0 14.18 38.7 308.8 0.531 3.39
1011 0.3 576,1728 3,4 105 2.686 582.5 36.43 17.1 611.2 0.478 1.89
1012 0.16 576,2304 3,4 70 4.067 1109 90.97 8.93 1397 0.477 1.12
No-slip 7 700 109 0.7 256,384 3,3 320 1.255 71.63 3.941 53.7 72.61 0.637 6.77
1010 0.5 384,768 3,3 160 1.535 146.4 10.11 40.9 152.4 0.616 3.69
1011 0.32 480,960 3,4 200 1.935 299.1 26.12 25.1 304.8 0.504 1.89
1012 0.16 768,2304 2,4 160 2.756 602.2 66.60 16.7 596.5 0.459 1.05
The resolutions in the y-direction are the same as those in the x-direction. The global density ratio is fixed at Λ = 1.6

larger in the two regions above and below the middle finger usually have smaller Le, into the content of the real environ-
layer, where d f is almost constant. ments with larger Le, such as the oceans.
We also calculate d f at different heights for the other Specifically, we found that the dependence of concentra-
three cases with RaC = 1012 , which are included in Fig. 9 tion flux on the concentration Rayleigh number is affected
for comparison. Clearly, only for the case with (Pr , Sc) = by Pr and Sc more strongly than those of heat flux and
(7, 70) the finger width d f near the boundary is much higher flow Reynolds number. The behaviors of the heat flux depend
than that in the middle of the bulk, indicating different layers more on Le, i.e. the ratio between Sc and Pr , instead of the
in the flow domain. d f only varies slightly for all the other exact values of Pr and Sc. In our set up, two thin boundary
three cases, corresponding to a nearly homogeneous finger layers exist adjacent to the top and bottom plates, and a fin-
layer in the bulk. These results provide limited but valuable ger layer develops in the bulk between two boundary layers.
information about the parameter ranges where the staircase If one measures the local properties of the finger layer, i.e.
occurs. The lower bound of RaC for the staircase regime is those defined by the local mean gradients of scalar compo-
about 1012 for (Pr , Sc) = (7, 70), or Le = 10. The other two nents, similarities can be observed between the results of our
combinations of (Pr , Sc) both have Le = 100, and no sign configuration and those from fully periodic domains.
of the staircase state emerges. Therefore, the minimal RaC Although the global density ratio is fixed in all our simu-
for the staircase to happen increases as Le becomes larger. lations, the local density ratio Λ f of the middle finger layer
For Le = 100 as in the ocean, the minimal RaC is beyond varies for different global parameters. Λ f decreases as RaC
1012 , which is very challenging for numerical simulation. increases, but all the values are much larger than those typ-
ically found in the ocean, for which a much higher RaC is
needed. The dependence of the density flux ratio γ f on Λ f
4 Conclusions and discussions in the finger layer is very close to those fund in the fully peri-
odic simulations. The horizontal width of the fingers scales
−1/4
In the present work we numerically investigate the fingering as d f ∼ RaC , which is consistent with the wavelength of
DDC which is driven by the concentration difference and sta- the fastest growing modes.
bilized by the temperature difference. Especially, we examine Finally, we observe that the staircase state starts to emerge
the effects of different combinations of molecular diffusiv- in only one case of our simulations, saying (Pr , Sc) =
ities. Three different combinations of Prandtl and Schmidt (7, 70) and RaC = 1012 . Fingering staircase is one of the
numbers are considered. Such results not only provide new most intriguing phenomena caused by DDC in the ocean, and
insights into the dynamics of fingering DDC, but also help us many questions remain open in the field. The current results
to put the experimental and numerical observations, which provide useful informations about the parameter range where

123
Y. Yang

the staircase state may exist, which will be investigated in 13. Schmitt, R.W.: Double diffusion in oceanography. Annu. Rev. Fluid
future. Mech. 26, 255–285 (1994). https://doi.org/10.1146/annurev.fl.26.
010194.001351
14. Radko, T.: Double diffusive convection. Cambridge Uni-
Acknowledgements This work was supported by the Major Research
versity Press, Cambridge (2013). https://doi.org/10.1017/
Plan of National Natural and Science Foundation of China for Turbulent
CBO9781139034173
Structures (Grants 91852107 and 91752202).
15. Stern, M.E.: The salt-fountain and thermohaline convection. Tellus
12, 172–175 (1960). https://doi.org/10.1111/j.2153-3490.1960.
tb01295.x
Appendix 16. Turner, J.S.: Salt fingers across a density interface. Deep-
Sea Res. 14, 599–611 (1967). https://doi.org/10.1016/0011-
7471(67)90066-6
References 17. Linden, P.F.: On the structure of salt fingers. Deep-Sea Res. 20,
325–340 (1973). https://doi.org/10.1016/0011-7471(73)90057-0
1. Garaud, P.: Double-diffusive convection at low Prandtl number. 18. Krishnamurti, R.: Double-diffusive transport in laboratory thermo-
Annu. Rev. Fluid Mech. 50, 275–298 (2017). https://doi.org/10. haline staircases. J. Fluid Mech. 483, 287–314 (2003). https://doi.
1146/annurev-fluid-122316-045234 org/10.1017/S0022112003004166
2. Debras, F., Chabrier, G.: New models of Jupiter in the context of 19. Hage, E., Tilgner, A.: High Rayleigh number convection with dou-
Juno and Galileo. Astrophys. J. 872, 100 (2019). https://doi.org/ ble diffusive fingers. Phys. Fluids 22, 076603 (2010). https://doi.
10.3847/1538-4357/AAFF65 org/10.1063/1.3464158
3. Davies, C.J., Pozzo, M., Gubbins, D., Alfe, D.: Transfer of oxygen 20. Stellmach, S., Traxler, A., Garaud, P., et al.: Dynamics of finger-
to Earth’s core from a long-lived magma ocean. Earth Planet Sci. ing convection. Part 2 The formation of thermohaline staircases.
Lett. 538, 116208 (2020). https://doi.org/10.1016/J.EPSL.2020. J. Fluid Mech. 677, 554–571 (2011). https://doi.org/10.1017/jfm.
116208 2011.99
4. Reali, J.F., Garaud, P., Alsinan, A., Meiburg, E.: Layer formation 21. Paparella, F., von Hardenberg, J.: Clustering of salt fingers in
in sedimentary fingering convection. J. Fluid Mech. 816, 268–305 double-diffusive convection leads to staircase like stratification.
(2017). https://doi.org/10.1017/jfm.2017.26 Phys. Rev. Lett. 109, 014502 (2012). https://doi.org/10.1103/
5. Xue, N., Khodaparast, S., Zhu, L., et al.: Laboratory layered latte. PhysRevLett.109.014502
Nat. Commun. 8, 1960 (2017). https://doi.org/10.1038/s41467- 22. Traxler, A., Stellmach, S., Garaud, P., et al.: Dynamics of fingering
017-01852-2 convection. Part 1 Small-scale fluxes and large-scale instabilities.
6. Tait, T., Howe, M.: Thermohaline staircase. Nature 231, 178–179 J. Fluid Mech. 677, 530–553 (2011). https://doi.org/10.1017/jfm.
(1971). https://doi.org/10.1038/231178a0 2011.98
7. Schmitt, R.W., Ledwell, J.R., Montgomery, E.T., et al.: Enhanced 23. Yang, Y., Verzicco, R., Lohse, D.: Scaling laws and flow structures
diapycnal mixing by salt fingers in the thermocline of the tropical of double diffusive convection in the finger regime. J. Fluid Mech.
Atlantic. Science 308, 685–688 (2005). https://doi.org/10.1126/ 802, 667–689 (2016). https://doi.org/10.1017/jfm.2016.484
science.1108678 24. Yang, Y., Verzicco, R., Lohse, D.: Vertically bounded double dif-
8. Correa-Ramirez, M., Rodriguez-Santana, A., Ricaurte-Villota, C., fusive convection in the finger regime: comparing no-slip versus
Paramo, J.: The Southern Caribbean upwelling system off Colom- free-slip boundary conditions. Phys. Rev. Lett. 117, 184501 (2016).
bia: water masses and mixing processes. Deep Sea Res. I 155, https://doi.org/10.1103/PhysRevLett.117.184501
103145 (2019). https://doi.org/10.1016/J.DSR.2019.103145 25. Ostilla-Monico, R., Yang, Y., van der Poel, E.P., et al.: A multiple
9. Durante, S., Schroeder, K., Mazzei, L., et al.: Permanent thermo- resolutions strategy for direct numerical simulation of scalar turbu-
haline staircases in the Tyrrhenian Sea. Geophys. Res. Lett. 46, lence. J. Comput. Phys. 301, 308–321 (2015). https://doi.org/10.
1562–1570 (2019). https://doi.org/10.1029/2018GL081747 1016/j.jcp.2015.08.031
10. You, Y.: A global ocean climatological atlas of the Turner angle: 26. Kunze, E.: A review of oceanic salt-fingering theory. Prog.
implications for double-diffusion and water-mass structure. Deep Oceanogr. 56, 399–417 (2003). https://doi.org/10.1016/S0079-
Sea Res. 49, 2075–2093 (2002). https://doi.org/10.1016/S0967- 6611(03)00027-2
0637(02)00099-7 27. Shraiman, B.I., Siggia, E.D.: Heat transport in high-Rayleigh-
11. Johnson, G.C., Kearney, K.A.: Ocean climate change fingerprints number convection. Phys. Rev. A 42, 3650–3653 (1990). https://
attenuated by salt fingering? Geophys. Res. Lett. 36, L21603 doi.org/10.1103/PhysRevA.42.3650
(2009). https://doi.org/10.1029/2009GL040697
12. Turner, J.S.: Multicomponent convection. Annu. Rev. Fluid Mech.
17, 11–44 (1985). https://doi.org/10.1146/annurev.fl.17.010185.
000303

123

You might also like