Access the full text.
Sign up today, get DeepDyve free for 14 days.
Loading next page...
/lp/arxiv-cornell-university/penetration-of-a-cooling-convective-layer-into-a-stably-stratified-otXqAAVBqr
References
References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.
- ISSN
- 2469-990X
- eISSN
- ARCH-3341
- DOI
- 10.1103/PhysRevFluids.5.124501
- Publisher site
- See Article on Publisher Site
Abstract
Penetration of a cooling convective layer into a stably-stratified composition gradient: entrainment at low Prandtl number J. R. Fuentes and A. Cumming Department of Physics and McGill Space Institute, McGill University, 3600 rue University, Montreal, QC H3A 2T8, Canada We study the formation and inward propagation of a convective layer when a stably-stratified fluid with a composition gradient is cooled from above. We perform a series of two-dimensional simulations using the Bousinessq approximation with Prandtl number ranging from Pr = 0:1 to 7, extending previous work on salty water to low Pr. We show that the evolution of the convection zone is well-described by an entrainment prescrip- tion in which a fixed fraction of the kinetic energy of convective motions is used to mix fluid at the interface with the stable layer. We measure the entrainment eciency and find that it grows with decreasing Prandtl number or increased applied heat flux. The kinetic energy flux that determines the entrainment rate is a small fraction of the thermal energy flux carried by convective motions. In this time-dependent situation, the density ratio at the interface is driven to a narrow range that depends on the value of Pr, and with low enough values that advection dominates the interfacial transport. We characterize the interfacial flux ratio and how it depends on the interface stability. We present an analytic model that accounts for the growth of the convective layer with two parameters, the entrainment eciency and the interfacial heat transport, both of which can be measured from the simulations. I. INTRODUCTION riors (Pr = = 0:001–0:1) have become possible. This work shows that while thermo-composional convective layers can also exist at low Pr [15–18], there are fundamental dif- In astrophysics, there are many situations in which a con- ferences in how and whether layers form and the nature of vective zone coexists with a composition gradient. A classic doubly-di usive convection [18]. The reader is referred to example is the convective core of a massive star, which is im- the excellent review by Garaud [19] for further details. These mersed in a gradient of heavy elements that results from nu- simulations have guided new transport prescriptions that can clear burning [1–4]. Gas giant planets, which undergo strong be included in 1D evolution codes [18]. Conditions in stellar convection in their gaseous envelopes, might develop compo- interiors, where Pr . 10 , are still inaccessible numerically. sition gradients from either their formation history, or colli- Despite the progress in understanding layer formation when sions during their evolution [5, 6]. Recently, the Juno mission there are pre-existing temperature and composition gradients, [7] has found evidence that an extended region of Jupiter’s less attention has been paid to situations in which the large- interior is enriched in heavy elements [8, 9]. scale gradients develop over time. An example is the penetra- The nature of convective mixing in these regions is not tion of a convective region into a neighbouring stable region clear. It is well known that composition gradients tend to sta- with a composition gradient. This configuration is relevant bilize the fluid against overturning convection [10], but the in the evolution of gas giant planets, in which a convective resulting transport of heat and heavy elements is not well- zone propagates inwards as the planet cools down, enriching understood. In stellar evolution, mixing across the bound- its outer regions by transporting heavy elements from below ary between a convection zone and a stable region can be ex- [e.g., 20, 21]. In this context, there are two relevant ques- tremely important because it can bring fresh fuel for nuclear tions: 1) how quickly does the outer convective layer move reactions into the convection zone. Evolution models for as- inwards, and 2) does the fluid become fully-mixed? In the trophysical objects over long timescales rely on analytic pre- context of Jupiter, for example, recent 1D evolutionary mod- scriptions for transport both within the convection zone and at els find that global composition gradients can persist over long the boundary. These are typically based on mixing length the- timescales, by separating into a number of distinct convective ory [e.g., 11, 12] and then implemented into one-dimensional layers, although not over as extensive a region as inferred from numerical models [e.g., 13]. the Juno data [20, 21]. These simulations, however, lack a de- Observations and numerical simulations of geophysical flu- tailed model of how composition and heat are transported at ids with composition gradients have shown that under cer- convective boundaries. tain circumstances, double-di usive instabilities lead to a se- ries of convective layers. The layers are well-mixed in both Several laboratory studies have been carried out in which composition and temperature, but separated by sharp inter- stably-stratified salty water is heated from below, creating faces across which transport of heat and composition is by a convective region that penetrates into the stably-stratified molecular di usion [e.g., 14, and references therein]. Astro- layer [22–25]. Motivated by experimental results, Turner [23] physical fluids di er in a key aspect, that the Prandtl num- developed a simple analytical model for the growth of the con- ber Pr = = , which measures the ratio of kinematic vis- vective layer. The fluid is assumed to be initially isothermal cosity to thermal di usivity , is Pr < 1 as opposed to with a linear salinity gradient dS=dz < 0, and a constant heat Pr 7 for salty water. Recently, with the improvement of flux F is applied at the bottom boundary. The model as- computational resources, three-dimensional numerical simu- sumes that at the top of the well-mixed convection zone there lations at low Prandtl numbers appropriate for planetary inte- is an abrupt step of both temperature T and salinity S (i.e., arXiv:2007.04265v4 [astro-ph.SR] 28 Jun 2021 2 molecular di usion of heat and salt are ignored). After a time the top with a constant heat flux. Our simulations were per- t, when the convective zone has a thickness h, from heat and formed with Pr ranging from 0.1 to 7 (i.e. we also include the salinity balance it follows that salty water regime for comparison), at fixed di usivity ratio = = 0:1 (with the solute di usivity). S T S c Th = F t ; (1) 0 P 0 The paper is organised as follows. In Sect. II we describe 1 dS the physical model and the numerical code used to perform S = h ; (2) the simulations. Sect. III presents a description of the in- 2 dz wards propagation of the convective layer. In Sect. IV we where is a background density and c is the specific heat 0 P measure the entrainment eciency at small Pr. In Sect. V we at constant pressure. The rate at which the convection zone discuss the relevance of the heat flux across the interface be- grows depends on the stability of the interface, S= T tween the convection zone and the stable layer, and its e ect R , where and are the coecients of solute contraction and on the growth of the layer. In Sect. VI we discuss the relative thermal expansion (both assumed to be positive constants). sizes of heat and composition transport across the interface. For a given value of R , eqs. (1) and (2) give In Sect. VII we present an analytic model of the evolution of the convective layer that reproduces our numerical results. Fi- ! ! 1=2 1=2 1=2 F dS 1=2 nally, we conclude in Sect. VIII. h(t) = 2R t : (3) c dz 0 P Turner [23] considered two limits for R . One possibility is II. MODEL AND NUMERICAL METHOD that the convection zone grows by Rayleigh-Taylor instabili- ties, when its temperature has increased enough to lower the We perform two-dimensional simulations in a horizontally- density jump at the interface to 0, ie. R 1. How- periodic domain of height H and width L. We impose im- ever, additional mixing mechanisms could in principle trans- permeable and stress-free top and bottom boundaries with no port heat and salt across a Rayleigh-Taylor stable interface, composition flux through them, no heat flux at the bottom, leading to a more rapid growth of the convective layer. For and a constant heat flux at the top. We use the Boussinesq ap- example, Kelvin-Helmholtz instabilities at the boundary can proximation [29], valid for a thin layer of fluid in which fluc- lead to entrainment of fluid from the stable layer [25]. As a tuations in density () are small with respect to the constant limiting case, Turner [23] found R = 3 under the assumption background density ( ). The density variations depend on that the potential energy change from heating the convective temperature and solute perturbations (T and S , respectively) layer is used to redistribute the heavy elements. via = ( S T ), where and are the coecients of Both experimental and numerical results for salty water solute contraction and thermal expansion, respectively, both suggest that entrainment at the interface does in fact play a key assumed to be positive constants. The governing equations role. While the initial measurements of Turner [23] suggested are that R 1, later experiments by Fernando [25] showed that the density interface is stable with a non-negligible buoyancy r v = 0 ; (4) jump across it (R > 1 or B g= < 0). Fernando [25] @T proposed that mixing across the stable interface occurs due to = (vr) T + r T ; (5) @t shear motions near the interface, and predicted the same rela- @S tion as in Eq. (3) for the growth rate of the convective layer, = (vr) S + r S ; (6) @t but with R replaced by a di erent constant that depends on @v rP the entrainment eciency. In an attempt to clarify the discrep- 2 = (vr) v + g + r v ; (7) ancy between Turner [23] and Fernando [25], Molemaker and @t 0 0 Dijkstra [26] performed two-dimensional numerical simula- with boundary conditions tions, with a similar set-up as in the classic laboratory exper- @u @S iments but cooled from above instead of heated from below. w = 0 ; = 0 ; = 0 ; (8) z=0;H Their results agreed with Fernando [25], giving support to en- @z @z z=0;H z=0;H trainment as the mixing mechanism. They also found that dif- @T @T F = 0 ; = : (9) fusive heat flux through the interface is significant, modifying @z @z k z=0 z=H Eq. (3). In the above equations, v = (u; w) is the velocity of a fluid ele- In this work, we investigate how low Pr a ects the growth ment, where u is the x-component, and w is the z-component, of a convective layer into a composition gradient. While P denotes the pressure fluctuation resulting from the motion of there has been some work done with a time-dependent back- the fluid, g is the acceleration due to gravity, and k = c ground temperature profile at low Pr [27, 28], it was fo- 0 P T is the thermal conductivity. Further, F corresponds to the cused on the formation and evolution of layers. Here we fo- 0 constant heat flux at the top boundary that cools the domain. cus on the physics behind the growth of the convective zone. The fluid is initialized with constant temperature T every- In particular, we investigate the eciency of entrainment at 0 lower Pr numbers. To accomplish this, we perform a se- where and with a linear composition profile S (z) = S + 0 0 ries of two-dimensional numerical experiments of an incom- S (1 z=H), with S defined such that the solute concentra- 0 0 pressible fluid with a linear composition gradient, cooled from tion is larger by a factor of two at the bottom of the domain. 3 we show them normalized to relevant reference values. For TABLE I. Parameters used in the simulations. example, the thickness of the convective layer is presented in Parameter Value terms of the height of the box (H), time is presented in terms H Height (m) 0.25 of the thermal di usion time across the box (t = H = ), di T L Width (m) 0.25 and temperature and solute are presented in terms of the ini- 7 2 1 Kinematic viscosity (10 m s ) 0:142, 1:42, 10 tial temperature and initial solute contrast across the box (T 7 2 1 Thermal di usivity (10 m s ) 1:42 and S , respectively). Further, the heat fluxes are presented T 0 7 2 1 in terms of F , and solute fluxes in terms of the initial solute Solute di usivity (10 m s ) 0:142 0 1 1 flux across the box ( jdS =dzj). For the interested reader, k Thermal conductivity (W m K ) 0.6 0 S 0 we present in Sect. II A a set of dimensionless equations with Background density (kg m ) 1025 the relevant dimensionless parameters that control our simu- 1 1 c Specific heat capacity (J K kg ) 4182 lations. 1 4 Thermal expansion coecient (K ) 2:3 10 Solute contraction coecient (1) 7:6 10 T Background temperature (K) 293.15 A. Dimensionless Parameters S Background solute (g kg ) 12.78 S Initial solute contrast across depth (g kg ) 13 In the following, we non-dimensionalize the Boussinesq F Critical heat flux for stability (W m ) 103 crit equations presented above such that length is in units of the F Heat flux at the top boundary (W m ) 5.4F , 10.8F 0 crit crit box height (H), time is in units of the thermal di usion time across the box (H = ), solute is units of the initial solute con- trast across the box (S ), and temperature is in units of the Afterwards, the fluid is destabilized by a constant heat flux F imposed flux as F H=k. By these choices, velocity is in units at the top boundary that drives the evolution of the system in 2 2 of =H, and pressure has units of =H . The resulting T 0 time. We choose the magnitude of F in terms of the di u- dimensionless equations are sive heat flux that would be present in the fluid if it was just marginally stable against convection r v ˜ = 0 ; (11) dS S 0 0 @T F = k = k ; (10) crit ˜ ˜ = (v ˜ r) T +r T ; (12) dz H @t @S i.e., we set F = f F , where f is a positive number (5.4 2 0 crit ˜ ˜ = (v ˜ r) S + r S ; (13) @t and 10.8 in our numerical experiments). The parameter values 2 3 6 7 used in the simulations were chosen to reproduce the experi- @v ˜ 6 F 7 6 7 2 ˜ 6 ˜ ˜ 7 = (v ˜ r) v ˜ rP +R Pr T S z ˆ + Prr v ˜ ; 6 7 4 5 ments of Turner and Stommel [22] and are shown in Table I. @t F crit Note that the solute di usivity was increased by an order (14) of magnitude such that = 0:1, and the kinematic viscosity was varied to have a set of simulations that covers Pr = 0:1, with boundary conditions 1, and 7. @u ˜ @S Since we are interested in the early evolution of the system, w ˜ = 0 ; = 0 ; = 0 ; (15) z ˜=0;1 @z ˜ @z ˜ z ˜=0;1 z ˜=0;1 our numerical experiments were performed until t 4500 s ˜ ˜ (i.e., t = 0:01 t , where t is the thermal di usion time @T @T di di = 0 ; = 1 : (16) across the box). This is enough time to observe the forma- @z ˜ @z ˜ z ˜=0 z ˜=1 tion of the outer convective layer and its inwards propagation We clarify that dimensionless variables are written with before the formation of secondary layers. a tilde and they should not be confused with horizontally- We solve linear terms implicitly and nonlinear terms explic- averaged (x-independent) variables, which are written with a itly using an implicit-explicit (IMEX), third-order, four-stage line on the top. Runge-Kutta time-stepping scheme RK443 with the Dedalus The dimensionless parameters that control the simulations spectral code [30]. The variables are decomposed on a Cheby- are F =F , the Prandtl number (Pr), the di usivity ratio ( ) 0 crit shev (vertical) and Fourier (horizontally-periodic) domain in and a modified Rayleigh number (R ), defined respectively as which the physical grid dimensions are 3/2 the number of modes. Based on a resolution study, we find that 512 modes F S 0 0 = F k ; (17) in each direction is enough to resolve all the fluid flows given F H crit the parameters used in this work. However, for a better res- Pr = ; (18) olution of small scale structures, we use 1024 modes in each direction. = ; (19) Although we solve the equations in dimensional form, most of the relevant parameters analysed and presented in this work gH F H are dimensionless. Further, for a better interpretation of the re- R = : (20) sults, when plotting the quantities that are not dimensionless, T 4 profiles for the case Pr = 0:1 and F = 5:4F at t = 2280 s 0 crit TABLE II. Dimensionless parameters used in the simulations. (t = 0:005 t ), the same snapshot as shown in Fig. 1. Despite di 7 2 1 # Pr F =F R R (10 m s ) 0 crit T S the fluctuations due to the advective contribution to the fluxes, 12 11 1 0.1 0.1 5.4 4 10 7:5 10 0:142 it is clear that in the convective layer the total heat flux in- 12 11 2 0.1 0.1 10.8 8 10 7:5 10 0:142 creases linearly with depth (Fig. 2a), meaning that the fluid is 11 10 cooling everywhere at a constant rate to keep its temperature 3 0.1 1 5.4 4 10 7:5 10 1:42 11 10 uniform. A similar behaviour is observed in the composition 4 0.1 1 10.8 8 10 7:5 10 1:42 10 10 flux (Fig. 2b). In the convective zone the total flux decreases 5 0.1 7 5.4 5:76 10 1:06 10 10 linearly with depth, thereby, the solute content is increasing 11 10 6 0.1 7 10.8 1:15 10 1:06 10 10 everywhere at the same rate to keep the fluid with uniform composition. Note that R (F =F ) can be re-written as Figure 3 shows the evolution in time of the thickness of the T 0 crit convective zone. To help compare the di erent simulations, we remove the h / F =F scaling predicted by Turner’s ! 0 crit 3 p F gH S 0 0 analytic model (Eq. 3) by plotting h=(H F =F ). For com- 0 crit R = R = ; (21) p T S crit T parison, we show h=(H F =F ) as predicted by Eq. (3) us- 0 crit ing R = 1 and R = 3. Comparing the di erent curves, we which looks as the traditional Rayleigh-number but for solute. see that there is a weak dependence of the rate of growth of the In terms of the non-dimensionalization described above, the convection zone on Pr, such that the convective layer grows parameters used in our set of six simulations are given in Table faster as Pr decreases. For example, at t = 0:01 t , the height di II of the convective zone at Pr = 0:1 is larger than for Pr = 7 It is worth mentioning that in this problem convection by a factor of two. Comparing curves at the same Pr, we is driven by the temperature di erence across the thermal see also that the growth rate of the convective layer increases boundary layer due to the imposed heat flux at the top, and slightly faster with flux than the expected F =F scaling. 0 crit the convective layer grows in time. This means that within the This can be seen in Fig. 3 where the curves for F = 10:8F 0 crit convection zone, the classic Rayleigh number and Reynolds lie slightly above those for F = 5:4F . The maximum devi- 0 crit number have a time-dependent magnitude determined by the ations between the curves for di erent fluxes are 2:8%; 6:4%, thickness of the convective layer and 13:2%, for Pr = 0:1; 1 and 7, respectively. As we discuss below, the variations with Pr can be understood in terms of 3 di erences in the entrainment eciency with Pr, as well as gh T v h Ra = ; Re = ; (22) the e ect of the heat flux at the boundary between the convec- tion zone and stable layer, which is not included when deriv- ing Eq. (3). where v is the convective velocity. By measuring T , h, and v at each time directly from the simulations, we find Ra and c The best-fit power law to the convection zone depth as func- 1=2 Re varying from 0 (initially) until a maximum value of 10 , tion of time is close to but not exactly h / t . Fitting a gen- 0:467(5) 0:585(2) and 10 , respectively. eral power law to the data, we find h / t , where the lowest and highest rate correspond to the cases (Pr = 7, F = 5:4F ), and (Pr = 0:1, F = 10:8F ), respectively 0 crit 0 crit III. INWARDS PROPAGATION OF THE CONVECTIVE (the values in parenthesis correspond to the uncertainties in LAYER the last digit). For Pr = 7, [25] and [26] found that their data 0:36 0:5 was fit by h / t depending on the magnitude of the We find that the initial behaviour of the system is qualita- imposed flux F . tively similar for all the simulations: after turning on the heat flux at the top, the cooling rate is high enough that a convec- tive layer, well mixed in both temperature and composition, quickly forms and grows inwards by incorporating fluid from IV. ENTRAINMENT AT THE CONVECTIVE BOUNDARY below, as shown in the snapshots in Fig. 1. To get some intuition on how temperature and composi- tion change within the convective layer, we look into the In this section, we investigate entrainment at the convec- horizontally-averaged profiles of heat and solute fluxes, which tive boundary as the mechanism responsible for mixing and we define as growth of the convective layer at Pr 1. In particular, we show that: 1) during the propagation of the convective layer, F = c wT k dT=dz (23) H 0 P a buoyancy jump across the interface is present, which sug- gests that a process is needed to transport heavier fluid across F = wS dS =dz ; (24) S 0 0 S the stable interface; 2) the entrainment equation proposed and respectively. The first and second term on the right hand side tested by Fernando [25] and Molemaker and Dijkstra [26] in in Eqs. (23) and (24) correspond to the advective and di usive experiments and simulations of salty water (Pr = 7) gives a fluxes, respectively. As an example, we show in Fig. 2 the flux good description of our results at lower Pr. 5 FIG. 1. Instantaneous snapshots of the temperature field (divided by the initial temperature T , left panel) and solute field (divided by the initial solute contrast S , right panel) for the case Pr = 0:1 and F = 5:4F , at t = 0:005 t . Blue (red) color represents low (high) temperature. 0 0 crit di Dark (light) color represents low (high) solute concentration. Note how convective eddies impinging on the interface incorporate fluid from below. 1.0 Advection Convective Advection Convective zone Diffusion Diffusion zone Total Total 0.8 (b) (a) 0.6 0.4 0.2 0.0 0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0 z/H z/H FIG. 2. Horizontally-averaged flux profiles for the simulation Pr = 0:1 and F = 5:4F at t = 2280 s (t = 0:005 t ). Panels (a) and (b) show 0 crit di profiles of heat and solute flux, respectively. In both panels, the green, blue, and orange lines correspond to the total, advective, and di usive contribution to the flux, respectively. The shaded areas denote the extent of the convective zone. Note that in panel (a) the heat fluxes are normalized to F , and in panel (b) the solute fluxes are normalized to jdS =dzj. Further, in both panels the z coordinate is normalized to 0 0 S 0 the height of the box (H). F /F H 0 F /( |dS /dz|) S 0 S 0 6 0.35 The buoyancy jump B also exhibits a linear trend with h, but Turner (R = 3) Pr = 1, F = 10.8F ρ 0 crit its magnitude is larger for simulations at Pr = 0:1. It is inter- Pr = 1, F = 5.4F Turner (R = 1) 0 crit esting that the ratiojBj=g S increases slowly with h, being 0.30 Pr = 7, F = 10.8F Pr = 0.1, F = 10.8F 0 crit 0 crit roughly constant for each experiment. We clarify that roughly Pr = 7, F = 5.4F Pr = 0.1, F = 5.4F 0 crit 0 crit constant means maximum variations at the level of 20%. We 0.25 find that the solute di erence across the interface accounts for 20-80% of the buoyancy jump, depending on Pr and F . For 0.20 comparison, Fernando [25] and Molemaker and Dijkstra [26] found for salty water that the salinity jump across the inter- 0.15 face accounts for 11% and 50% of the buoyancy jump, respec- tively. The di erences can be explained by the magnitude of the imposed heat flux and the initial solute gradient. In terms 0.10 of our units, Fernando [25] and Molemaker and Dijkstra [26] used F 18F and F 5:6F , respectively. 0 crit 0 crit 0.05 0.00 B. Entrainment equation and mixing eciency 0.000 0.002 0.004 0.006 0.008 0.010 t/t diff A key parameter of the entrainment mechanism is the so- called mixing eciency. The entrainment hypothesis states FIG. 3. Temporal evolution of the thickness of the outer convective that the rate of change of potential energy due to mixing is layer, h, divided by H F =F . The gray and black dashed-lines 0 crit proportional to the kinetic energy flux available near the in- correspond to the predictions by Eq. (3) using R = 1 and 3, respec- terface [e.g., 31]. By assuming that a constant fraction of the tively. Colors distinguish between di erent Pr, and line-style distin- guish between di erent F (dotted-lines in the case F = 10:8F , 0 0 crit available kinetic energy is used to lift heavier fluid across the and solid-lines for F = 5:4F . 0 crit interface, Fernando [25] and Molemaker and Dijkstra [26] de- rived an expression for the rate of change of the convection zone thickness A. Development of a buoyancy jump in a stable interface ( B) dh F = ; (25) g dt c 0 P Figure 4 shows horizontally-averaged profiles of the tem- perature, T , composition, S , and buoyancy, B = g( T S ), which defines the mixing eciency, [see, e.g., the dis- at di erent times for the case Pr = 0:1 and F = 5:4F . A 0 crit cussion in Sect. 3.2.4 in 26]. The entrainment rate is of- buoyancy jump at the base of the convection zone develops ten also written in terms of a bulk Richardson number Ri = and persists over time. We found the same behavior in all our hB=v , where v is the rms convective velocity and we use simulations. To show this more clearly, we show in the inset the height of the convective layer as the length-scale of the of panel (c), the profile at a particular time, with the region turbulent motions. Using mixing-length theory to write v denoting the buoyancy jump using a thicker red line. 1=3 1=3 (g F = c ) h , Eq. (25) takes the form dh=dt v =Ri. 0 0 P c Figure 5 shows the jumps in solute, temperature, and buoy- We find 10 . Ri . 100 for all our numerical experiments. ancy across the interface (S , T , and B, respectively) as Our results fall within the same parameter range reported a function of the thickness of the convective layer, h. We in previous laboratory experiments of turbulent entrainment measure the jump in each quantity from horizontally-averaged [32, 33], and hydrodynamics simulations of stellar convective profiles, defined as the value below the interface (stable re- boundaries [34]. This corresponds to the intermediate stability gion) minus the value above the interface (convective region), regime in which the convective zone expands and the interface so that T and S are positive quantities, whereas B is is moderately distorted by convective eddies. For much larger negative for a stable interface. It is worth mentioning that values of Ri the entrainment process weakens and the evolu- the dispersion in our measurements is due to the propaga- tion of the interface is expected to be controlled by di usive tion of waves near the interface, which make its location (start processes [25–27]. and end) time-variable, especially in the simulated experiment Fernando [25] and Molemaker and Dijkstra [26] found in with Pr = 0:1 and F = 10:8F . 0 crit their experiments at Pr = 7 that increases slowly with time, We observe that the jumps in solute, temperature and buoy- with maximum variations at the level of 30%. They reported ancy all exhibit a monotonic (positive) trend with h, weakly time-averaged values of between 0.15-0.56 depending on dependent on F . As expected, since solute is conserved dur- the magnitude of the imposed heat flux at the boundary. In the ing the evolution of the convective layer, S exhibits a linear following, we test whether exhibits a similar behaviour at lower Pr. trend with h (Eq. 2), independent of Pr and F . The situation We compute at di erent times by using the buoyancy for T is less clear and there are substantial di erences be- tween the simulations, probably due to the e ect of heat flux jumps B from horizontally-averaged profiles, as the ones in at the interface between the convective layer and stable region. Fig. 4, and dh=dt from di erentiation of a power law fit to the h/H F /F 0 crit 7 1.00 1.00 1.00 0.90 0.90 0.90 0.80 0.80 0.80 0.70 0.70 0.70 0.60 0.60 0.60 0.50 0.50 0.50 Time Time Time 0.40 0.40 0.40 (a) (b) (c) 0.30 0.30 0.30 0.985 0.990 0.995 1.000 1.0 1.1 1.2 1.3 1.4 0.053 0.054 0.055 0.056 0.057 T/T S/δS 0 0 B/g FIG. 4. Horizontally-averaged profiles of the temperature (normalized to T ), solute (normalized to S ), and buoyancy field (normalized to 0 0 the magnitude of the acceleration due to gravity, g), in panels a, b, and c, respectively. The results corresponds to the simulation at Pr = 0:1 and F = 5:4F . Profiles are shown at di erent times (where time evolves according the direction of the arrows). All panels share the same 0 crit scale along the z-axis. Note that the z coordinates is hormalized to the height of the box, H. In panel (c), the buoyancy profile at t = 0:005 t di is shown in red and a zoomed-region shows the buoyancy step using a thicker line. curves h(t) in Fig. 3. Despite the dispersion due to measure- V. THE EFFECT OF THE INTERFACIAL HEAT FLUX ment uncertainties in B, the evolution of behaves similarly at low and high Pr, increasing slowly with time, with max- For Pr = 7, Molemaker and Dijkstra [26] pointed out that imum variations at the level of 20-40 % (Fig. 6). We find there is a significant heat flux across the interface between the that the time-averaged values of take values between 0:08 convection zone and stable layer below. This has the e ect of and 1, being higher at low Pr and high F . The trend with heating the convective layer from below and thereby reducing F is less clear at Pr = 0:1 since the flow is more turbulent the rate at which it penetrates into the stable layer. In this sec- and the dispersion in the measurements is higher. Our results tion, we present our measurements of the interfacial heat flux make sense given that a larger value of F provides more en- as the convective layer evolves, and test whether it is signifi- ergy to the convective eddies, thereby they can entrain and cant at low Pr. mix more eciently. Furthermore, low Pr fluids have strong The change in the heat content within the convective layer velocity gradients near the interface, enhancing shear motions of thickness h [26] is determined by and mixing. Finally, low Pr fluids are more turbulent (have a larger Reynolds number) and deliver energy to smaller scales dT i c h = F F , (26) with the result that entrainment might be expected to be more 0 P 0 dt ecient. Note that decreasing Pr at a fixed thermal di usiv- ity means that the thickness of the viscous boundary layer that i where F = c Tdh=dt + F is the total heat flux through 0 P a separates the convective layer and the static fluid below gets the interface. The term c Tdh=dt corresponds to heat flux 0 P smaller, thereby convective eddies entrain through a thinner through the interface that results from a change dh = hdt in layer, mixing the fluid more easily. the thickness of the convective layer, and F is additional heat Our results compare reasonably well with previous work. flux from below. Note that with F = 0, Eq. (26) reduces to Our measurements of for simulations at Pr = 7 ( 0:08 for F = 5:4F , and 0:12 for F = 10:8F ) are expected to Eq. (1). We measure F from the flux profiles in Fig 2(a) as 0 crit 0 crit be smaller than those reported in Fernando [25], who obtained the value of the total heat flux at the edge of the convective 0:5 for F 18F . However, for the case F = 5:4F , zone. 0 crit 0 crit we expected consistency with Molemaker and Dijkstra [26], Figure 7 shows for all our simulations the temporal evolu- who obtained 0:15 for F 5:6F in Molemaker and 0 crit tion of the total heat flux through the interface, F , normal- Dijkstra [26], but our measurement is roughly smaller by a ized to the imposed cooling flux F . For comparison, we also factor of 2. include the contribution of the c Tdh=dt term. Interest- 0 P At this point, we have shown that during the propagation ingly, we find that F is weakly-dependent of Pr and F , and of the convective layer, a buoyancy jump develops over the it fluctuates around a constant value 0:6F . The contribu- interface. Further, using the entrainment equation (Eq. 25), tion from c Th also fluctuates around a constant value but 0 P we have shown that behaves in a similar way at low and high it is slightly di erent depending on F and Pr. We subtract Pr, increasing slowly with time. We have also shown that is c Th from F , and take the temporal average between higher at low Pr and high F , which suggests that entrainment 0 P 0 H is stronger in the more turbulent and energetic flow. 1000 4500 s to quantify F for all our simulations. z/H 8 Pr = 0.1, F = 5.4F Pr = 1, F = 10.8F 0 crit 0 crit Δ S = 0.5 h|dS /dz| 0.35 0 0.045 Pr = 0.1, F = 10.8F Pr = 7, F = 5.4F 0 crit 0 crit Pr = 0.1, F = 5.4F 0 crit Pr = 1, F = 5.4F Pr = 7, F = 10.8F 0 crit 0 crit Pr = 0.1, F = 10.8F 0 crit 0.040 0.30 Pr = 1, F = 5.4F 0 crit (b) Pr = 1, F = 10.8F 0 crit 0.035 Pr = 7, F = 5.4F 0.25 0 crit Pr = 7, F = 10.8F 0 crit 0.030 0.20 (a) Δ 0.025 0.15 0.020 0.10 0.015 0.05 0.010 0.00 0.0 0.2 0.4 0.6 0.0 0.2 0.4 0.6 h/H h/H 1.0 Pr = 0.1, F = 5.4F Pr = 1, F = 10.8F Pr = 0.1, F = 5.4F 0 crit 0 crit 0 crit 0.0025 Pr = 0.1, F = 10.8F Pr = 7, F = 5.4F Pr = 0.1, F = 10.8F 0 crit 0 crit 0 crit 0.9 Pr = 1, F = 5.4F Pr = 7, F = 10.8F 0 crit 0 crit Pr = 1, F = 5.4F 0 crit Pr = 1, F = 10.8F 0 crit 0.8 (d) 0.0020 Pr = 7, F = 5.4F 0 crit Pr = 7, F = 10.8F 0 crit 0.7 (c) 0.6 0.0015 β 0.5 0.0010 0.4 0.3 0.0005 0.2 0.1 0.0 0.2 0.4 0.6 0.0 0.2 0.4 0.6 h/H h/H FIG. 5. Jumps at the interface as a function of h=H (the thickness of the convective layer normalized to the height of the box). Panels (a), (b), and (c) show the absolute jumps of composition (normalized to S ), temperature (normalized to T ), and buoyancy (normalized to the 0 0 magnitude of the acceleration due to gravity, g), respectively. Panel (d) shows the ratiojBj=g S versus h=H. As shown in the legends, colors distinguish between di erent Pr, and markers distinguish between di erent values of F We find that F is a fixed fraction of the imposed heat flux VI. BUOYANCY TRANSPORT ACROSS THE INTERFACE at the top, F = "F , with " varying between 0:25 and 0:5, a 0 therefore it significantly a ects the growth rate of the convec- tion zone (Fig. 8). Further, we observe that F increases with The buoyancy jump at the bottom of the convective layer Pr and for all the simulations at F = 5:4F , it is 25% 0 crit suggests that there must be a net transport of buoyancy across larger than for F = 10:8F . This result makes sense because 0 crit the interface as the convection zone grows. In this section we at high Pr the thickness of the convective layer is smaller, investigate the relative heat and solute fluxes at the interface. thereby the temperature of the convective layer drops more First, similarly to the heat flux in Sect. V, we measured the quickly. This implies a higher temperature contrast with the solute flux at the interface. This is shown in the left panel of fluid below (Fig. 5b), resulting in more di usion of heat up- Fig. 9. We find that the solute flux agrees well with flux im- wards. The fact that increased F slows the convection zone plied by the growth rate of the layer, S h. We also observe growth is consistent with the curves of h(t)=H F =F in 0 crit that the solute transport is higher at low Pr and high F , con- Fig. 3, which show that for F = 10:8F the curves lie above 0 crit sistent with the fact that the convective layer grows faster in the ones for F = 5:4F , and the di erence between them 0 crit these cases. All of these results are consistent with and ex- increases from low to high Pr. pected from mass conservation. An indication of the nature of the transport at the interface | B|/g S/δS T/T | B|/(g S) 9 1.4 Experimentally, di erent relations between R and R Pr = 0.1, F = 5.4F Pr = 1, F = 10.8F 0 crit 0 crit have been reported for the transport across a single interface Pr = 0.1, F = 10.8F Pr = 7, F = 5.4F 0 crit 0 crit bounded by two convective layers in salty water (rather than 1.2 Pr = 1, F = 5.4F Pr = 7, F = 10.8F 0 crit 0 crit an interface between a convective layer and a stable layer as we study here). Turner [36] found that for 2 < R < 7, the 1.0 flux ratio R is a constant, independent of R . This was con- firmed by Linden and Shirtcli e [37] who found that the value 1=2 1=2 of R was consistent with R = ( = ) = . Further F F S T 0.8 analysis by Newell [38] showed that at very large R , the flux ratio obeys Eq. (29). The di erences in the behaviour of R 0.6 were attributed to the nature of the transport across the in- terface. At low R , advection dominates the fluxes and en- hances the transport of salt, whereas at large R , the transport 0.4 is dominated by molecular di usion. Note that in the latter 1=2 case (transport by di usion), both relations R = R and 0.2 R = R have succeeded at explaining di erent experimental data [38, 39]. More recently, in the context of the transport of heavy elements between the core and the gaseous envelope of 0.002 0.004 0.006 0.008 0.010 t/t Jupiter, Moll et al. [40] performed three-dimensional simula- diff tions for Pr = = 0:03 0:3. They identified the advective and di usive regimes of the interface described above, but in FIG. 6. Entrainment parameter as a function of t=t (i.e., time di both regimes the buoyancy flux ratio was roughly independent normalized to the thermal di usion time across the box) for all 1=2 of R and significantly greater than . our numerical simulations. As shown in the legends, colors distin- guish between di erent Pr, and markers distinguish between di er- Fig. 10 shows our measurements of the buoyancy flux ra- ent F . For simulations at F = 5:4F , the averaged values of for 0 0 crit tio as a function of R . We find that R increases with R , so Pr = (0:1; 1; 7) are (0:85; 0:44; 0:08), whereas for simulations that as the convection zone deepens and the interface becomes at F = 10:8F , the averaged values are (0:84; 0:51; 0:12). 0 crit more stable (larger R ), there is a larger solute flux compared to heat flux. As expected, since the total heat flux through the interface is approximately the same for all our experiments, is the relation between the buoyancy flux ratio we find that the evolution of R scales in the same way as the solute flux F , i.e., R decreases with increasing Pr and in- R (27) creases with increasing F . The range of values of R seems F 0 F 1=2 F c to converge towards as Pr increases, consistent with the H P measurements for Pr = 7 in laboratory experiments. How- and the stability of the interface characterized by the density 1=2 ever, in all cases we find that R > , consistent with the ratio parameter, R S = T (defined here such that R > results in Moll et al. [40], although our values of R are signif- 1 indicates a stable interface). For example, if the transport icantly larger than theirs, as shown in Fig. 10c. In particular, were only by di usion in the interface, the solute and heat we find for the cases Pr = = 0:1 that R 1:5 2, whereas fluxes are given by Moll et al. [40] found R 0:7. 0 1 0 1 Also shown in Fig. 10 are the values of R computed us- B C B C i S i T B C B C B C B C F B C ; F c B C ; (28) ing the di usive fluxes of solute and heat only. In this case, 0 S @ A 0 P T @ A S H S T the values are consistent with R = R , as in the laboratory F I experiments by Newell [38]. As mentioned above, this im- where and are the thicknesses of the di usive boundary S T plies that the di usive boundary layers of solute and temper- layers of solute and temperature, respectively. If , this S T ature have the same thickness. Indeed, direct measurement of gives the boundary layer thicknesses confirms this, and is shown in Fig. 11. R = R : (29) However, it might be expected that and would have a S T di erent thickness. Fernando [35] suggested that the inter- face thickness is set by a balance between the di usion time VII. ANALYTIC MODEL FOR THE INWARDS across the layer and the convective turnover time. Using mix- PROPAGATION OF THE CONVECTIVE LAYER ing length theory for the convective flux and equating it to the di usive flux across the layer gives The fact that the excess heat flux across the interface F is 1=2 a fixed fraction of the imposed heat flux, F = "F (section R = R (30) a 0 V), and that the entrainment parameter varies slowly in time instead. (section IV), suggest the following set of equations to describe 10 Pr = 0.1, F = 10.8F , Total Pr = 1, F = 10.8F , Total Pr = 7, F = 10.8F , Total 0 crit 0 crit 0 crit Pr = 0.1, F = 5.4F , Total Pr = 1, F = 5.4F , Total Pr = 7, F = 5.4F , Total 1.0 0 crit 1.0 0 crit 1.0 0 crit ˙ ˙ ˙ Pr = 0.1, F = 10.8F , ρ c Δ Th Pr = 1, F = 10.5F , ρ c Δ Th Pr = 7, F = 10.8F , ρ c Δ Th 0 crit 0 p 0 crit 0 p 0 crit 0 p ˙ ˙ ˙ Pr = 0.1, F = 5.4F , ρ c Δ Th Pr = 1, F = 5.4F , ρ c Δ Th Pr = 7, F = 5.4F , ρ c Δ Th 0 crit 0 p 0 crit 0 p 0 crit 0 p 0.8 0.8 0.8 0.6 0.6 0.6 0.4 0.4 0.4 0.2 0.2 0.2 0.002 0.004 0.006 0.008 0.010 0.002 0.004 0.006 0.008 0.010 0.002 0.004 0.006 0.008 0.010 t/t t/t t/t diff diff diff FIG. 7. Heat flux through the interface (normalized to F ) as a function of t=t (time normalized to the thermal di usion time across the 0 di box). As shown in the legends, colors distinguish between di erent Pr, and markers distinguish between di erent F . All panels share the same scale in both axes. In all panels, filled and unfilled colors distinguish between the total heat flux and the resulting flux due to a change dh = hdt in the convective zone, respectively 0.55 where B = g T S . This extends the analytic mod- els of Turner [23] and Fernando [25] to include both entrain- ment and the heat flux across the interface. 0.50 It is worth noting that there is a separation of energy scales in this problem that allows us to write the global energy bal- 0.45 ance in Eq. (31) separately from the energy considerations that lead to the entrainment equation (33). The energy required to 0.40 3 mix the heavy elements, E = g dS =dz H =12 per unit mix 0 0 area [23], is a small fraction of the total thermal energy lost 0.35 by the layer, Pr = 0.1, F = 5.4.4F 0 crit 0 1 Pr = 0.1, F = 10.8F 0 crit B C E 1 S gH gH mix B C 0.30 7 B C B C = 10 ; (34) Pr = 1, F = 5.4F 0 crit @ A 6 c c P P c T H T 0 P Pr = 1, F = 10.8F 0 crit 0.25 Pr = 7, F = 5.4F 0 crit where we write S = H dS =dz =2. Using mixing length es- Pr = 7, F = 10.8F 0 crit timates F v c T and v gH T (where v is a conv P conv conv 0.20 0 2 4 6 typical convective velocity and T a typical temperature fluc- Pr tuation in the convection zone), we see that the kinetic energy flux F associated with the convective motions is smaller KE FIG. 8. Ratio F =F as a function of Pr. As shown in the legends, a 0 than the thermal energy carried by convection by the same colors distinguish between di erent Pr, and markers distinguish be- factor, tween di erent F (for a given Pr). The error bars correspond to one 3 2 sigma from the mean. F v v gH KE conv conv : (35) F v c T c T c 0 conv P P P Eq. (33) describes how this much smaller component of the the location of the interface: energy, the kinetic energy,is used to entrain heavy fluid and move it across the interface. These contributions to the en- dT dh F ergy, however, are only small corrections to the overall ther- h = T + (1 ") ; (31) dt dt c 0 P mal energy balance described by Eq. (31). We now explore the consequences of this model. For sim- 1 dS S = h ; (32) plicity and to get an analytic solution, we assume " and con- 2 dz ! stants (this choice is justified by the fact that both quantities dh g F vary slowly with time, with maximum variations at the level of B = ; (33) dt c 0 P less than 40%). The set of equations (31)–(33) has a solution F /F (Interface) ε = F /F a 0 11 200 200 200 Pr = 0.1, F = 10.8F Total 0 crit Pr = 1, F = 10.8F Total Pr = 7, F = 10.8F Total 0 crit 0 crit Pr = 0.1, F = 10.8F , ρ Δ Sh 0 crit 0 ˙ ˙ Pr = 1, F = 10.8F , ρ Δ Sh Pr = 7, F = 10.8F , ρ Δ Sh 0 crit 0 0 crit 0 175 Pr = 0.1, F = 5.4F Total 175 175 0 crit Pr = 1, F = 5.4F , Total Pr = 7, F = 5.4F Total 0 crit 0 crit Pr = 0.1, F = 5.4F , ρ Δ Sh 0 crit 0 ˙ ˙ Pr = 1, F = 5.4F , ρ Δ Sh Pr = 7, F = 5.4F , ρ Δ Sh 0 crit 0 0 crit 0 150 150 150 125 125 125 100 100 100 75 75 75 50 50 50 25 25 25 0.002 0.004 0.006 0.008 0.010 0.002 0.004 0.006 0.008 0.010 0.002 0.004 0.006 0.008 0.010 t/t t/t t/t diff diff diff FIG. 9. Solute flux through the interface (normalized to the initial di usive solute flux, jdS =dzj) as a function of t=t (i.e., time 0 S 0 di normalized to the thermal di usion time across the box). As shown in the legends, colors distinguish between di erent Pr, and markers distinguish between di erent F . In all panels, filled and unfilled colors distinguish between the total heat flux and the resulting flux due to a change dh = hdt in the convective zone, respectively. 4.0 4.0 4.0 Pr = 0.1, F = 5.4F , Total Pr = 0.1, Total Pr = 7, Diffusion Pr = 0.1, Total Pr = 7, Diffusion 0 crit 1/2 1/2 Pr = 0.1,F = 10.8F , Total Pr = 1, Total τ Pr = 1, Total τ 0 crit 3.5 3.5 3.5 1/2 τR τR Pr = 7, Total ρ Pr = 7, Total ρ τ 1/2 1/2 Pr = 0.1, Diffusion Pr = 0.1, Diffusion Moll et al. (2017) τ R τ R ρ ρ 3.0 3.0 3.0 Pr = 1, Diffusion Pr = 1, Diffusion 2.5 2.5 2.5 (a) F = 5.4F (b) F = 10.8F 0 crit 0 crit (c) 2.0 2.0 2.0 1.5 1.5 1.5 1.0 1.0 1.0 0.5 0.5 0.5 0.0 0.0 0.0 1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 R R R ρ ρ ρ FIG. 10. Ratio between the buoyancy fluxes of solute and temperature, R , as a function of the density ratio parameter, R . Panels (a) and (b) show the results for F = 5:4F and F = 10:8F , respectively. Panel (c) shows the results for the simulations at Pr = = 0:1, and 0 crit 0 crit the results reported by Moll et al. [40]. As shown in the legends, colors distinguish between di erent Pr, and markers distinguish between di erent F as well as di erent flux ratios (filled markers consider the total flux, whereas unfilled markers consider just the di usion flux). The black lines in all panels corresponds to di erent predictions (see text for more details). 1=2 h / t which is the constant C is identified as R which must be larger than unity (since an interface mixes by Rayleigh-Taylor instability 1=2 1=2 1=2 as soon as it reaches R = 1), so the fact that we infer C = 0:75 ( ) ( ) h( ; "; t) = 2C( ; ") t ; (36) crit for Pr = 7 implies that additional physics must be at work. We can also use Eqs. (31)–(33) and the solution Eq. (36) to where F is given by Eq. (10). This is the same as Eq. (3) crit calculate the fluxes at the interface, and derive the expected but with a di erent prefactor. The constant C is given in terms relation between R and R . First, Eqs. (32), (33), and (36) of the parameters and " as give C( ; ") = 1 " + 2 : (37) B 2 = ; (38) Note that and " can be measured directly from the simula- 1 " + 2 g S tions: see Fig. 6 for and Fig. 8 for ". For example, taking S 1 " + 2 0:85 (0:1) and " 0:3 (0:45) gives C 2:4 (0:75) for R = = : (39) 1 " Pr = 0:1 (7). Note that in the original model by Turner [23], T F /( |dS /dz|) 0 S 0 S 12 1.4 The buoyancy flux ratio is Pr = 0.1, F = 5.4F 0 crit i ! ! Pr = 1, F = 5.4F 0 crit 1.3 F 1 " + 2 1 " Pr = 7, F = 5.4F 0 crit R = = = R ; (43) 1 1 + " 1 + " Pr = 0.1, F = 10.8F 0 crit F c 1.2 Pr = 1, F = 10.8F 0 crit Pr = 7, F = 10.8F 0 crit which increases with R as observed. 1.1 We compare the model predictions and the measurements from the simulations in more detail in Fig. (12). By using 1.0 the temporal averages and standard deviations of and ", we propagate their errors to get the uncertainties in the predictions above. We find that within the uncertainties, there is a good 0.9 agreement between the model predictions and our numerical results. 0.8 0.7 VIII. SUMMARY AND CONCLUSIONS 0.002 0.004 0.006 0.008 0.010 We studied the penetration of a cooling convection zone t/t diff into a stably-stratified composition gradient at low Pr. Our goal was to extend previous work on salty water at Prandlt FIG. 11. Ratio between the thickness of the di usive boundary layer number Pr 7 to low values Pr < 1 found in planetary inte- of solute and temperature ( = ), as a function of t=t . As shown S T di riors. Our main conclusions are: in the legends, colors distinguish between di erent Pr, and markers distinguish between di erent F . 1. A non-negligible buoyancy jump develops over the in- terface between the convective layer and the stratified region (Fig. 5c). The stability of the interface as mea- The first of these explains the ratio B=g S found in Fig. 5 sured by the density ratio R = S = T increases (panel d). To the extent that and " vary slowly in time, so is slowly with with time as the convective layer grows, the stability of the interface, which is determined by the values with a value ranging between 1 . R . 4 depending on of and ". Again taking 0:85 (0:1) and " 0:3 (0:45) Prandtl number. for Pr = 0:1 (7), we find R = 3:4 (1:4) and B=g S = R 1:0 = 0:3 ( 0:7) (compare Figs. 5 and 10). 2. Our results are well-described by an entrainment pre- Eq. (39) shows that the range of values of R depends on the scription in which a fixed fraction of the kinetic en- maximum value of . The definition of in eq. (33) suggests ergy associated with the convective motions is used to that should not be much larger than unity, since in that case lift heavier fluid across the interface, as proposed by the energy required to mix fluid across the interface would ex- Fernando [25] and Molemaker and Dijkstra [26] for ceed the available kinetic energy. With " = 0, R = 1 + 2 , salty water. The entrainment eciency (Eq. [25]) is which has a value R = 3 when = 1. This matches Turner’s approximately constant in time (with variations at the argument [23] based on energetics for the maximum stabil- level of 20 - 40 %). This confirms and extends to lower ity of the interface. When the heat flux across the interface Pr previous work identifying entrainment as the mixing is included, larger values of R are possible, as seen in our mechanism responsible for the growth of the outer con- simulations. For example, for the Pr = 0:1 value " = 0:45, vective layer rather than Rayleigh-Taylor instabilities. R 4:6 for = 1. The continued cooling of the convection 3. Entrainment is stronger at low Pr and high imposed flux zone continuously destabilizes the interface, preventing large F . This implies that mixing is more ecient when the values of R . The constant C in eq. (36) can be rewritten flow is more turbulent and energetic, with the result that the convective layer grows more quickly in those cases C( ; ") = R (1 "); (40) (Fig. 3). The entrainment parameter changes from 0:1 at Pr = 7 to 0:9 at Pr = 0:1, so while entrainment so we see that compared to Turner’s estimate in eq. (36), the is a relatively minor e ect at Pr = 7, it is much more height of the interface at a given time is smaller by a factor significant at low Pr. 1=2 (1 ") . Eqs. (31), (32), and (36), also give expressions for the total 4. As pointed out previously by Molemaker and Dijkstra flux of solute and heat through the interface [26], additional interfacial heat flux, presumably asso- ! ! ! ciated with the transport of solute across the interface, i F 1 " + 2 F = S h = ; (41) is a significant fraction of the imposed heat flux at the c 2 top boundary (see Fig. 8). The flow of energy into the convective layer reduces the e ective cooling rate of the F = c Th + "F = (1 + ") : (42) 0 P 0 2 convection zone. /δ S T 13 0.200 200 Pr = 0.1, F = 10.8F Pr = 0.1, F = 10.8F Pr = 0.1, F = 10.8F 0 crit 0 crit 0 crit 1.0 0.175 175 Pr = 1, F = 10.8F Pr = 1, F = 10.8F Pr = 1, F = 10.8F 0 crit 0 crit 0 crit Pr = 7, F = 10.8F Pr = 7, F = 10.8F Pr = 7, F = 10.8F 0 crit 0 crit 0 crit 0.150 150 (c) 0.8 (b) (a) 0.125 125 β 0.6 0.100 100 0.075 75 0.4 0.050 50 0.2 0.025 25 0.000 0.002 0.004 0.006 0.008 0.010 0.002 0.004 0.006 0.008 0.010 0.002 0.004 0.006 0.008 0.010 t/t t/t t/t diff diff diff Pr = 0.1, F = 10.8F Pr = 0.1, F = 10.8F 0 crit 0 crit 1.0 Pr = 1, F = 10.8F Pr = 1, F = 10.8F 0 crit 0 crit 2.5 Pr = 7, F = 10.8F Pr = 7, F = 10.8F 0 crit 0 crit (d) 0.8 (e) 2.0 0.6 1.5 0.4 1.0 0.2 0.5 0.002 0.004 0.006 0.008 0.010 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 t/t R diff ρ FIG. 12. A comparison of the analytic model given by Eqs. (31)–(33) with the results of our simulations. Panels (a), (b), (c), and (d) show for Pr = (0.1, 1, 7), and F = 10:8F , the temporal evolution (normalized to thermal di usion time, t ) of h=H, the ratio jBj=g S , the solute 0 crit di flux through the interface normalized to the initial di usive solute flux, F =( jdS =dzj), and the heat flux through the interface normalized 0 S 0 to F , F =F , respectively. Panel (e) shows the relation between the buoyancy flux ratio, R , and the density ratio parameter, R . In all panels 0 H 0 F the shaded regions represent the predictions from the model by considering 1-sigma uncertainties in and . As shown in the legends, colors distinguish between di erent Pr 5. We find that the interfacial heat and composition fluxes Our focus in this paper has been on the growth of the outer are dominated by advection rather than di usion (this convection zone, with the goal of addressing how low Pr af- can be seen in Fig. 2). Because the stability of the in- fects the rate at which it moves into the stably-stratified re- terface is limited to R . 3–5 (depending on Pr), it gion. Another important question is whether secondary layers is always in a regime where advection dominates the develop, slowing the progress of the convective region, and in interfacial transport. The interface adjusts so that the principle preventing the system from mixing fully. Secondary thickness of the temperature and salinity boundary lay- layers are seen in salt water experiments, but it is not known ers are the same to 10%, despite the fact that the when and how they arise in time-dependent cooling at low Pr. molecular di usivities are di erent by a factor of ten In this regard, a few attempts have been made [27, 28]. Biello ( = = = 0:1). [27] found that gravity waves can break near the interface and S T mix the composition gradient across, making the formation of secondary layers dicult to occur at low Pr. On the other 6. Equations (31)–(33) provide a simple analytical model hand, Zaussinger and Kupka [28] found that multiple layers that reproduces our numerical results with two param- can form at low Pr either by a thermal instability at the in- eters (assumed constant): the entrainment eciency terface ahead of the main convective layer, or spontaneously (Fig. 6), and the heat flux across the interface as a frac- develop due to double-di usive instabilities, as the ones ob- tion of the applied heat flux at the top of the convection served in Radko [14] and Mirouh et al. [16]. We will discuss zone " = F =F (Fig. 8). The growth of the convection a 0 these issues in a companion paper. 1=2 zone thickness follows h / t and is given by Eq. (36). Eq. (39) gives R in terms of and ". We based our simulations on the pioneering salt-water ex- h/H F /F R | B|/(g S) F /( dS /dz|) S 0 S 0 14 periments of Turner and Stommel [22], reducing the fluid vis- tion with 1 (note that as in salty water is also 1 in cosity to lower the Prandtl number. The lowest value of Pr Jupiter’s interior [47]). In stellar convection, the distinction we consider, Pr = 0:1, is at the upper end of values expected is less important since there the equation of state is closer to to occur in planetary interiors, where Prandlt numbers may ideal gas with 1; still the kinetic energy flux can be as extend down to 10 . In stellar interiors, even lower val- small as 0:01 of the total heat flux (see discussion in [34]). ues Pr 10 are expected. Our results suggest that the en- We have made a number of approximations which should trainment rate may be near maximum already at Pr = 0:1, be relaxed in future work. Although two-dimensional simula- since 1, implying that a large fraction of the available ki- tions have been successful at reproducing the classic labora- netic energy is taken up by entrainment. Recent calculations tory experiments by Turner [36] and Fernando [25] (e.g. see of convective boundary mixing in stars also find entrainment Appendix A of [28]), it would be interesting to compare 3D rates that scale linearly with the convective flux [41–43], or in simulations with the same setup with our 2D results, particu- terms of bulk Richardson number as Ri [34, 44, 45], sup- larly at low Pr. Di erences between 3D and 2D may explain porting the kind of entrainment relation we have used here. the factor of 2 lower values of R measured at Pr = 0:1 by An interesting di erence is that in stars the composition dif- Moll et al. [40], although their interface was between two con- ference is produced internally by nuclear burning and so the vection zones rather than a convection zone and stable layer. interface can be a lot sti er than in our problem, where cool- In addition, in a planetary context, rotation and compressibil- ing of the convection zone quickly drives the stability of the ity are important (see [48] for a study of layer formation with interface R to smaller values R . 4. rotation at low Pr), and so simulations that go beyond the Even though entrainment at low Pr involves a substantial Boussinesq approximation and include rotation would be of fraction of the kinetic energy of convection, this energy is a great interest. small part of the overall energy budget (see Eqs. [34-35] and ACKNOWLEDGMENTS discussion in Sect. VII). The relevant energy is the kinetic en- ergy because ultimately shear instabilities at the interface mix the fluid; the fact that buoyancy drives convective motions We thank Pascale Garaud, Falk Herwig, Toby Wood, and means that the kinetic energy is naturally of the same scale Florian Zaussinger for useful discussions. This work was sup- as the energy required to overcome the buoyancy of the stable ported by an NSERC Discovery Grant. We also thank Ben interface. This is important for core erosion in Jupiter: Moll A. and Evan Anders for technical support with the Dedalus et al. [40] used the buoyancy flux ratio R from their simula- code. J.R.F. acknowledges support from a McGill Space In- tions to derive an expression for the core erosion rate that was stitute (MSI) Fellowship. A. C. and J. R. F. are members substantially smaller than the earlier suggestion based on the of the Centre de Recherche en Astrophysique du Quebec ´ total thermal flux integrated over the core radius [46]. The ra- (CRAQ) and the Institut de recherche sur les exoplanetes ` tio between the new erosion rate and the old rate is exactly the (iREx). This research was enabled in part by support provided ratio (Eq. [35]) between the kinetic energy in convection and by Calcul Quebec ´ (calculquebec.ca), and Compute Canada the thermal energy. The distinction between kinetic energy (www.computecanada.ca). Computations were performed on flux and heat flux is an important one in Boussinesq convec- Graham and Belug ´ a. [1] P. Ledoux, “Stellar Models with Convection and with Discon- N. Movshovitz, Y. Kaspi, R. Helled, D. Reese, E. Galanti, tinuity of the Mean Molecular Weight,” Astrophys. J. 105, 305 S. Levin, J. E. Connerney, and S. J. Bolton, “Comparing Jupiter (1947). interior structure models to Juno gravity measurements and [2] R. J. Tayler, “Evolution of Massive Stars.” Astrophys. J. 120, the role of a dilute core,” Geophys. Res. Lett. 44, 4649–4659 332 (1954). (2017). [3] M. Schwarzschild and R. Harm, ¨ “Evolution of Very Massive [9] Florian Debras and Gilles Chabrier, “New Models of Jupiter in Stars.” Astrophys. J. 128, 348 (1958). the Context of Juno and Galileo,” Astrophys. J. 872, 100 (2019). [4] William J. Merryfield, “Hydrodynamics of Semiconvection,” [10] S. Kato, “Overstable Convection in a Medium Stratified in Astrophys. J. 444, 318 (1995). Mean Molecular Weight,” PASJ 18, 374 (1966). [5] G. Chabrier and I. Bara e, “Heat Transport in Giant [11] N. Langer, K. J. Fricke, and D. Sugimoto, “Semiconvective (Exo)planets: A New Perspective,” Astrophys. J. Lett. 661, di usion and energy transport,” Astron. Astrophys. 126, 207 L81–L84 (2007). (1983). [6] J. Leconte and G. Chabrier, “A new vision of giant planet interi- [12] H. C. Spruit, “Semiconvection: theory,” Astron. Astrophys. ors: Impact of double di usive convection,” Astron. Astrophys. 552, A76 (2013). 540, A20 (2012). [13] Bill Paxton, Lars Bildsten, Aaron Dotter, Falk Herwig, Pierre [7] S. J. Bolton, J. Lunine, D. Stevenson, J. E. P. Connerney, Lesa re, and Frank Timmes, “Modules for Experiments in S. Levin, T. C. Owen, F. Bagenal, D. Gautier, A. P. Ingersoll, Stellar Astrophysics (MESA),” Astrophys. J. S. 192, 3 (2011). G. S. Orton, T. Guillot, W. Hubbard, J. Bloxham, A. Coradini, [14] T. Radko, “A mechanism for layer formation in a double- S. K. Stephens, P. Mokashi, R. Thorne, and R. Thorpe, “The di usive fluid,” J. Fluid Mech 497, 365–380 (2003). Juno Mission,” Space Science Reviews 213, 5–37 (2017). [15] E. Rosenblum, P. Garaud, A. Traxler, and S. Stellmach, “Tur- [8] S. M. Wahl, W. B. Hubbard, B. Militzer, T. Guillot, Y. Miguel, bulent Mixing and Layer Formation in Double-di usive Con- 15 vection: Three-dimensional Numerical Simulations and The- [33] J. L. McGrath, H. J. S. Fernando, and J. C. R. Hunt, “Turbu- ory,” Astrophys. J. 731, 66 (2011). lence, waves and mixing at shear-free density interfaces. Part 2. [16] G. M. Mirouh, P. Garaud, S. Stellmach, A. L. Traxler, and T. S. Laboratory experiments,” J. Fluid Mech 347, 235–261 (1997). Wood, “A New Model for Mixing by Double-di usive Con- [34] Casey A. Meakin and David Arnett, “Turbulent Convection in vection (Semi-convection). I. The Conditions for Layer Forma- Stellar Interiors. I. Hydrodynamic Simulation,” Astrophys. J. tion,” Astrophys. J. 750, 61 (2012). 667, 448–475 (2007). [17] T. S. Wood, P. Garaud, and S. Stellmach, “A New Model for [35] Harindra J. S. Fernando, “Buoyancy transfer across a di usive Mixing by Double-di usive Convection (Semi-convection). II. interface,” Journal of Fluid Mechanics 209, 1–34 (1989). The Transport of Heat and Composition through Layers,” As- [36] J. S. Turner, “The coupled turbulent transports of salt and and trophys. J. 768, 157 (2013). heat across a sharp density interface,” International Journal of [18] R. Moll, P. Garaud, and S. Stellmach, “A New Model for Heat and Mass Transfer 8, 759–767 (1965). Mixing by Double-di usive Convection (Semi-convection). III. [37] P. F. Linden and T. G. L. Shirtcli e, “The di usive interface in Thermal and Compositional Transport through Non-layered double-di usive convection,” Journal of Fluid Mechanics 87, ODDC,” Astrophys. J. 823, 33 (2016). 417–432 (1978). [19] P. Garaud, “Double-Di usive Convection at Low Prandtl Num- [38] T. A. Newell, “Characteristics of a double-di usive interface at ber,” Annual Review of Fluid Mechanics 50, 275–298 (2018). high density stability ratios,” Journal of Fluid Mechanics 149, [20] Allona Vazan, Ravit Helled, and Tristan Guillot, “Jupiter’s evo- 385–401 (1984). lution with primordial composition gradients,” Astron. Astro- [39] J. S. Turner, T. G. Shirtcli e, and P. G. Brewer, “Elemental phys. 610, L14 (2018). Variations of Transport Coecients Across Density Interfaces [21] Simon Muller, ¨ Ravit Helled, and Andrew Cumming, “The in Multiple-di usive Systems,” Nature 228, 1083–1084 (1970). Challenge of Forming a Fuzzy Core in Jupiter,” arXiv e-prints , [40] R. Moll, P. Garaud, C. Mankovich, and J. J. Fortney, “Double- arXiv:2004.13534 (2020), arXiv:2004.13534 [astro-ph.EP]. di usive Erosion of the Core of Jupiter,” Astrophys. J. 849, 24 [22] J. S. Turner and H. Stommel, “A New Case of Convection in (2017). the Presence of Combined Vertical Salinity and Temperature [41] Paul R. Woodward, Falk Herwig, and Pei-Hung Lin, “Hydro- Gradients,” Proceedings of the National Academy of Science dynamic Simulations of H Entrainment at the Top of He-shell 52, 49–53 (1964). Flash Convection,” Astrophys. J. 798, 49 (2015). [23] J. S. Turner, “The behaviour of a stable salinity gradient heated [42] S. Jones, R. Andrassy, S. Sandalski, A. Davis, P. Woodward, from below,” J. Fluid Mech 33, 183–200 (1968). and F. Herwig, “Idealized hydrodynamic simulations of turbu- [24] Herbert E. Huppert and P. F. Linden, “On heating a stable salin- lent oxygen-burning shell convection in 4 geometry,” MNRAS ity gradient from below,” J. Fluid Mech 95, 431–464 (1979). 465, 2991–3010 (2017). [25] Harindra J. S. Fernando, “The formation of a layered structure [43] R. Andrassy, F. Herwig, P. Woodward, and C. Ritter, “3D when a stable salinity gradient is heated from below,” J. Fluid hydrodynamic simulations of C ingestion into a convective O Mech 182, 525–541 (1987). shell,” MNRAS 491, 972–992 (2020). [26] M. J. Molemaker and H. A. Dijkstra, “The formation and evo- [44] H. Kato and O. M. Phillips, “On the penetration of a turbulent lution of a di usive interface,” J. Fluid Mech 331, 199–229 layer into stratified fluid,” J. Fluid Mech 37, 643–655 (1969). (1997). [45] A. Cristini, R. Hirschi, C. Meakin, D. Arnett, C. Georgy, and [27] J. A. Biello, Layer formation in semiconvection, Ph.D. thesis, I. Walkington, “Dependence of convective boundary mixing on THE UNIVERSITY OF CHICAGO (2001). boundary properties and turbulence strength,” MNRAS 484, [28] Florian Zaussinger and Friedrich Kupka, “Layer formation in 4645–4664 (2019). double-di usive convection over resting and moving heated [46] Tristan Guillot, David J. Stevenson, William B. Hubbard, and plates,” Theoretical and Computational Fluid Dynamics 33, Didier Saumon, “The interior of Jupiter,” in Jupiter. The Planet, 383–409 (2019). Satellites and Magnetosphere, Vol. 1, edited by Fran Bagenal, [29] E. A. Spiegel and G. Veronis, “On the Boussinesq Approxima- Timothy E. Dowling, and William B. McKinnon (2004) pp. tion for a Compressible Fluid.” Astrophys. J. 131, 442 (1960). 35–57. [30] Keaton J. Burns, Geo rey M. Vasil, Je rey S. Oishi, Daniel [47] Martin French, Andreas Becker, Winfried Lorenzen, Nadine Lecoanet, and Benjamin P. Brown, “Dedalus: A flexible frame- Nettelmann, Mandy Bethkenhagen, Johannes Wicht, and work for numerical simulations with spectral methods,” Phys. Ronald Redmer, “Ab Initio Simulations for Material Properties Rev. Research 2, 023068 (2020). along the Jupiter Adiabat,” Astrophys. J. Suppl. Series 202, 5 [31] P. F. Linden, “The deepening of a mixed layer in a stratified (2012). fluid,” J. Fluid Mech 71, 385–405 (1975). [48] Ryan Moll and Pascale Garaud, “The E ect of Rotation on Os- [32] Harindra J. S. Fernando, “Turbulent mixing in stratified fluids,” cillatory Double-di usive Convection (Semiconvection),” As- Annual Review of Fluid Mechanics 23, 455–493 (1991). trophys. J. 834, 44 (2017).
Journal
Physics – arXiv (Cornell University)
Published: Jul 8, 2020