Home

Search

Collections

Journals

About

Contact us

My IOPscience

A comparative study of approaches for modeling RayleighTaylor turbulent mixing

This article has been downloaded from IOPscience. Please scroll down to see the full text article. 2010 Phys. Scr. 2010 014012 (http://iopscience.iop.org/1402-4896/2010/T142/014012) View the table of contents for this issue, or go to the journal homepage for more

Download details: IP Address: 129.31.235.38 The article was downloaded on 04/03/2013 at 10:42

Please note that terms and conditions apply.

IOP PUBLISHING Phys. Scr. T142 (2010) 014012 (13pp)

PHYSICA SCRIPTA doi:10.1088/0031-8949/2010/T142/014012

A comparative study of approaches for modeling RayleighTaylor turbulent mixing

Snezhana I Abarzhi and Robert Rosner
Department of Astronomy and Astrophysics, Physical Sciences Division, The University of Chicago, Chicago, IL 30337, USA E-mail: snezha@uchicago.edu and r-rosner@uchicago.edu

Received 20 September 2010 Accepted for publication 13 December 2010 Published 31 December 2010 Online at stacks.iop.org/PhysScr/T142/014012 Abstract This paper considers similarities and differences in the governing mechanisms and basic properties of RayleighTaylor turbulent mixing as discussed in recent theoretical and heuristic modeling studies and briey discusses how these mechanisms and properties may be explored in experiments and simulations. We were motivated by a number of stimulating questions, thoughtful comments and sagacious remarks by our colleagues and the anonymous referees, whose contribution to improving this work is warmly appreciated. PACS numbers: 47.20.k, 47.20.i, 47.54.+r, 52.35.Py, 52.57.2, 52.35.g, 94.05.Lk

1. Introduction
RayleighTaylor (RT) turbulent mixing is an extensive interfacial mixing process that develops when uids of different densities are accelerated against their density gradient [1, 2]. It governs a wide variety of natural phenomena from astrophysical to atomistic scales and plays an important role in technological applications [35]. Examples include inertial connement and magneto-inertial fusion, lightmaterial interaction, supernova explosions, ows in the atmosphere and oceans, and shockturbulence interaction in aerodynamics and free-space optical telecommunications [613]. RT mixing is a multi-scale, heterogeneous, anisotropic and statistically unsteady turbulent process with non-local interactions among the many scales [3, 4]. The development of RT turbulent mixing is usually associated with the conditions of strong gradients of pressure and density and may also include spatially varying and time-dependent acceleration, diffusion of species, heat release, chemical reactions, etc [113]. These conditions thus depart from the conditions under which one might expect canonical Kolmogorov turbulence to occur [4, 1416]. Capturing the mechanisms and properties of RT mixing can lead to a better understanding of realistic turbulent ows and further develop our intuition and the methods of controlling non-equilibrium process in nature and technology.
0031-8949/10/014012+13$30.00

The RT ows that arise in a variety of diverse applications nevertheless exhibit a number of similar features of their evolution [14, 1727]. The mixing starts to develop when the uid interface is slightly perturbed near its equilibrium state. The ow transitions from an initial stage, where the perturbation amplitude grows relatively quickly (e.g. exponentially in time), to a nonlinear stage where the growth rate decreases and the interface is transformed into a composition consisting of a large-scale coherent structure and small-scale irregular structures driven by shear, and then nally to a stage of turbulent mixing, whose dynamics is believed to be self-similar [14, 1727]. Over 100 RT-related papers are published per year in peer-reviewed mathematical, computational, scientic and engineering journals. However, our knowledge of this mixing process and progress toward fuller understanding are still limited. On the experimental side, a systematic study of RT ows is a challenging task [1728]. Due to the sensitivity and transient character of the dynamics, repeatable implementation of these ows in a laboratory environment require tighter than usual control of the macroscopic parameters and of the initial and boundary conditions. Statistical unsteadiness of turbulent mixing indicates the need to measure spatial and temporal distributions of the ow quantities [28]. Furthermore, these measurements should be carried out with relatively high accuracy and resolution, adequate dynamic range and data acquisition rate [3, 28].
1
2010 The Royal Swedish Academy of Sciences Printed in the UK

Phys. Scr. T142 (2010) 014012

S I Abarzhi and R Rosner

On the side of numerical simulations, modeling RT mixing is challenging also because of the highly singular nature of the instability (namely the rapid initial growth of small-scale structures) combined with the limited spatial dynamic range of exemplary simulations, even those which are based on state-of-the-art adaptive mesh renement computations and the use of leadership-class massively parallel computers. The numerical solutions are likely to be strongly dependent on the effects of unresolved small-scale structures (especially insofar as they affect local density contrasts) and the possible anomalous character of energy transport [2939]. On the side of theoretical analysis, the dynamics of RT ows is an intellectually challenging problem, as it has to balance numerous competing requirements and demands due attention to the multi-scale, highly nonlinear and non-local character of the mixing ows. If possible, we would like to identify universal asymptotic solutions and to establish whether memory of the initial conditions exists, to capture a certain degree of order of the mixing ow and to account for the noisy character of the turbulent dynamics and so forth. Our aim is to use physics-based considerations for preventing the analysis from becoming too mathematical and too empirical, and to help identify a set of robust parameters, which could in principle be precisely diagnosed in the observations ([3, 4, 28] and references therein). For a detailed discussion of the state-of-the-art methods in theoretical, experimental, numerical and computational studies of RT ows, see recent reviews [40]. This paper considers only one aspect of the multi-faceted RT mixing problem, specically the mechanisms and properties of RT mixing suggested by theoretical and heuristic modeling studies [4, 4161]. Our discussion will be physics-based and will attempt to synthesize knowledge of the problem from the perspectives of analysis and modeling, as it was elaborated by the turbulent mixing community.

2. Approaches for modeling RT turbulent mixing

2.1. Outline of empirical approaches An overview of experimental, numerical and theoretical studies [1739] suggests that RT ow is characterized by a large-scale structure, small-scale structures and extensive transport of mass, momentum and energy among the scales. The large-scale structure has two macroscopic length scales: in the direction of gravity g and the spatial the amplitude h period in the normal plane [3]. The horizontal scale is induced by the initial perturbation and/or by the mode of fastest growth with wavelength ( 2 /g )1/3 , where is the kinematic viscosity and g = |g|. The horizontal scale may increase if the ow is two-dimensional (2D) and the perturbation is broad-band and incoherent [47]. The is believed to grow self-similarly in the vertical scale h gt 2 , and it can be regarded as mixing regime with h an integral scale representing cumulative contributions of small-scale structures, which are produced by shear at the uid interface [5961]. Some other features are induced in the dynamics by diffusion, compressibility, stratication, nite-size domian, non-uniform acceleration and high energy density conditions [1739].
2

Since the time of rst hypotheses on the existence of a self-similar regime in RT ows [41, 62] and rst endeavors to observe it in the experiments and simulations [43, 45, 17, 27], tremendous effort has been put by researchers in empirical and theoretical modeling studies of the mixing dynamics, and signicant success has been achieved [4161]. In addition to addressing fundamental scientic questions, theories and models [4161] have a denitive practical importance: because of the sensitivity, statistical unsteadiness and transient character of the dynamics, experiments and simulations on RT mixing require solid benchmarks for analyzing and calibrating data sets [1739], even in the case when data calibration is performed by means of nonlinear spline interpolation with multiple adjustable parameters. Virtually all modeling efforts [4161] have tried to address the following questions. (i) What is the mechanism underlying self-similar mixing? (ii) What are the mixing ow properties? A detailed description of the former set of efforts can be found in [49]. Representative examples of these approaches are the viable works [43, 45] and the models presented in [47, 48]. The models [43, 44, 4749] suggested gt 2 that the growth of the horizontal scale varies as h and they scaled the coarsest vertical length scale of the ow = f ( A)gt 2 , where is a constant, f ( A) is a function as h of the Atwood number A = (h l )/(h + l ) and h(l) is the density of the heavy (light) uid. These interpolation models traditionally served for a calibration of experimental and numerical data sets and for further elaboration of the rst quantitative descriptions of RT ows [34, 49]. The set of modeling efforts related to the turbulence models presented in [41, 42, 5058] were initiated by the seminal works [41, 42, 50, 58]. These models considered RT mixing within the general content of turbulent ows and have served to underpin the development of fast and affordable numerical modeling techniques of the mixing process (e.g. [32, 34, 38, 58]). These models regarded RT mixing as an anisotropic turbulent ow with density uctuations, which are in turn driven by the velocity eld as in the case of passive scalar mixing. In order to interpret experimental and numerical data in terms of turbulent power laws [1416, 6366], the models [5057] somewhat modied the classical Kolmogorov theory, including the introduction of a virtual origin and a time scale for the transition to turbulence, gt 2 in Kolmogorovs the introduction of time dependence h invariants, etc. The principal difference between the turbulence models [41, 42, 5058] and the interpolation models [4349] lies in how the dynamics of the interface is handled. The models [4349] were developed to capture the dynamics of the fronta sharp interface between the immiscible uids. In contrast, the models [41, 42, 5059] considered turbulent mixing in a continuous approximation, presuming that the uids are miscible and can therefore diffuse into one another. For an early time evolution with the perturbation amplitude being signicantly smaller than the spatial period, h , the sharp-boundary approximation and the continuous approximation provide equivalent results [13]. and h To capture late-time dynamics, i.e. with h , the models [4158] applied a number of empirical hypotheses (to be discussed below) augmented with adjustable parameters,

Phys. Scr. T142 (2010) 014012

S I Abarzhi and R Rosner

and found some quantitative agreement with the observational data (which, as an important aside, spanned a relatively short dynamic range [1739]). Concurrent with the empirical modeling efforts, rigorous theoretical studies were extensively conducted (see [3, 4] and the review article [40] and references therein). These studies conrmed some of the empirical hypotheses [4157], and found a number of new qualitative and quantitative properties of turbulent mixing. It was shown, for instance, that the nonlinear dynamics of RT ows has an essentially non-local and multi-scale character, characterized by independent ) scales, contributions of the horizontal () and vertical (h and that in highly isotropic coherent structures, growth of horizontal scales may not occur [3, 4]. Recently developed momentum-based considerations accounted for these results and identied two distinct mechanisms for the development of the mixing process [5961]. Furthermore, it revealed new invariant, scaling and spectral properties of RT mixing ows and showed the departure of these properties from the canonical Kolmogorov scenario [15]. In the following, we compare different approaches [4161] for modeling RT turbulent mixing and outline some directions for future development of the experiments and simulations. 2.2. Conservation laws As in any physical process, RT mixing is governed by the conservation of mass, momentum and energy. In continuous approximation, in the uid bulk the conservation of mass and momentum has the form + v = 0, + (v ) v + g) + p + S = 0, (v (1)

strong gradients of the ow quantities. A self-consistent description of such an interface requires establishing new connections of continuum matter approximation to kinetic processes at microscopic scales as well as a better understanding of the interplay between Eulerian and Lagrangian descriptions in systems that are out of local thermodynamic equilibrium [39, 40]. 2.3. Comparative study of theories and models 2.3.1. Governing principles of the models. In order to obtain a rigorous theoretical description of RT mixing, one has to nd a solution for a mathematical problem of fundamental complexity. This problem includes a three-dimensional non-stationary system of nonlinear partial differential equations augmented with the initial conditions and with the boundary conditions, which are represented by a sub-set of nonlinear partial differential equations at a nonlinear freely evolving interface, and with the conditions at the boundaries of the outside domain. Asymptotic solutions for this mathematical problem are sensitive to the initial conditions and to the inuence of secondary instabilities and singularities, which may develop at the discontinuities, e.g. the uid interface. While the mixing process is observed to maintain certain features of coherence and order, associated primarily with the dynamics of large scales, it is yet a non-deterministic and statistically unsteady process, whose randomness results from the interaction of all the scales [3, 4, 14, 60]. The rst attempts to capture nonlinear evolution of the unstable mixing front were made by Davies and Taylor [2], Fermi and von Neuman [62], Layzer [68], Birkhoff and Carter [69] and Garabedian [70] via application of the collocation method, Lagrange dynamics, local spatial expansion and conformal mapping techniques. Nearly 30 years later, on the basis of theoretical solutions [6870] and experimental observations [2, 17], the model presented by Sharp and coworkers [43] put forward a hypothesis that an increase of the spatial period may lead to ow acceleration and thus trigger the transition from the nonlinear stage to accelerated mixing as = Agt 2 , h (3.1) where v is the characteristic velocity of the front with v |v|. Good agreement between the renormalization group analysis [43] and 2D front-tracking numerical simulations [44] was achieved. To describe the experimental [17] and numerical data [45] on RT mixing, another empirical model was proposed by Youngs [45], , v = Ag C ( A)v 2 /h (3.2) with C ( A) being a function of the Atwood number, distinct for bubbles and spikes [45]. Model (3.2) suggests that mixing development is induced by a broad-band initial perturbation. This model is known to provide an excellent data t with minimal use of free parameters [45, 49]. Yet another interpolation model [47, 48] initiated a merger campaign with the use of an empirical equation balancing the ow inertia, buoyancy and drag as ( A)v 2 )Area, (v Ag )Volume = (C
3

where , v and p are the density, velocity and pressure of the uid, respectively, S denotes terms induced by viscous stress and other effects, and the dot marks the partial derivative in time t [14]. For compressible uids, system (1) is augmented with the equation for energy transport and the equation of state [67]; for miscible uids, the equations for concentration are also incorporated [14]. For incompressible immiscible uids, the uid interface is a discontinuity, with = h H ( ) + l H () and v = vh H () + vl H ( ), where H is the Heaviside step function, is a scalar function on the coordinates and time with = 0 at the interface and h(l) and vh(l) are the density and velocity of the heavy (light) uid located in the region < 0 ( > 0) [3, 14]. If there is no mass ow across the interface, the pressure and normal component of velocity are continuous at the interface: ph | =0 = pl | =0 , | |, vl n| =0 = vh n| =0 = / (2.1)

g ,

gt 2

v gt ,

where n = /| | is the interface unit normal vector. In spatially extended uid systems the ow has no mass sources vh |z + = vl |z = 0 (2.2)

and can be periodic in the plane (x , y ) normal to the direction of gravity, z . The initial conditions at the interface and at the boundaries of the domain close the set of governing equations (1) and (2). For compressible and/or miscible uids, the uid interface is a region characterized by

(3.3)

Phys. Scr. T142 (2010) 014012

S I Abarzhi and R Rosner

( A) being a function of the Atwood number, distinct with C 2 and with from C ( A) and with Area 2 , Volume h gt 2 . This the single-scale mixing mechanism h model attempted to provide a universal characterization of RT mixing with sustained and time-dependent acceleration histories g = g (t ), as well as in the special case of the RichtmyerMeshkov mixing with g = 0 for both 3D and 2D ows. The model of [47, 48] relies extensively on adjustable parameters and agrees with the experimental and numerical data processed by means of nonlinear interpolation [49]. Applying turbulence modeling approaches for RT mixing was initiated by the works of Belenki and Fradkin [41] and Neuvazhaev [42], who derived a self-similar solution for the mixing dynamics from the energy balance relation (1/2)(v 2 / t ) + v 3 / L = L v2 , (3.4)

where 2 = g ( ln / z + g/c2 ) is the BruntVisl frequency, L is the characteristic length scale, is kinematic viscosity and c is the speed of sound. Presuming that the density of an incompressible uid obeys the equation / t = ( D / z )/ z with the turbulent diffusion , and integrating over coefcients D v L and L h the mixing zone by analogy with Prandlts theory of shear-induced mixing, the models [41, 42] found self-similar gt 2 in the cases of sustained and solutions with h time-dependent accelerations. These results were extensively used for the calibration of experimental and numerical data sets [24, 25, 27] and provided good agreement with data in a broad parameter regime. A more rigorous turbulent approach was suggested by Harlow and co-workers [50, 51] for a two-uid system. Ristorcelli and Clark [53] applied a similar approach for a single uid system and expanded the velocity and density elds in the vicinity of their mean values within Boussinesq approximation, presuming the validity of this approximation for miscible uids with similar densities A 1. To the rst order, the model [53] nds that vv / z = p / z Ag , / t + v / z = D 2 / z 2 , (3.5)

cases of RT and RichtmyerMeshkov mixing in the high and low energy density regimes. The importance of the models [5056] is that these models formulated the problem of RT mixing within the general context of turbulent ows. Furthermore, inspired by the pioneering numerical approaches of Nikiforov et al and viable works of Gauthier et al on the application of the k sub-grid-scale model to numerical simulations of accelerated and shock-driven ows [58], the turbulent diffusion models [41, 42, 5056] helped develop affordable numerical techniques for large-scale computations of RT mixing and for comparison of simulation results with the experiments [31, 32, 34, 35, 38]. To interpret experimental and numerical data on RT mixing in terms of turbulent power laws, Chertkov [57] attempted to extend Kolmogorov and BolgianoObukhov analyses for the case of RT mixing via a substitution of time dependence of the integral scale L h gt 2 in the invariants of canonical turbulent ows. This substitution suggested that the viscous scale should decay with time as t 1/4 in the 3D case and grow as t 1/8 in the 2D case [57]. Some quantitative agreements were reported between the models [4158] and the observations [1739]. Some qualitative features of the turbulent process remain unclear, such as the relatively ordered character of RT mixing ow at high Reynolds numbers [56]. To date, experiments and simulations have not provided a trustworthy guidance on whether the concepts of classical turbulence are applicable to /gt 2 and an accelerating RT ow and whether the scalings h 2 /gt are indeed universal. In order to explain the qualitative features of RT mixing and to connect empirical models to conservation laws (1) and (2) and to rigorous theoretical studies [3, 4], a momentum-based consideration was developed [4, 5961]. It suggests that in RT mixing ow, the specic momentum is gained due to buoyancy and is lost due to dissipation. The dynamics of a parcel of uid is governed by a balance per unit mass of the rate of momentum gain and the rate of momentum loss as = v, v h = . (3.6)

where v is the z -component of velocity, indicates the Favre averaging and is an effective concentration. As in [41, 42], equations (3.4) have the similarity solution = (C0 /4) Ag (t + t0 )2 with t0 , h 0 being the time scale and h length scale of the virtual origin, at which the transition to 0 = C0 Agt 2 turbulence is expected to occur and at which h 0 with C0 being a free parameter [53]. The authors of [53] noted that in accelerated mixing ow, the rate of energy dissipation is time dependent with g 2 t . The turbulent diffusion models [41, 42, 5053] agreed with experiments and simulations, similarly to the interpolation front-tracking models [4349]. The results presented in [53] were consistent with the nding of Dimotakis [54], who analyzed a wide variety of turbulent ows and suggested that at certain values of Reynolds and Taylor Reynolds numbers a transition may occur to a fully developed isotropic turbulence and that such a transition is a universal phenomenon for all turbulent ows including RT mixing. Zhou et al [55, 56] further extended Dimotakis ideas to the
4

The rate of momentum gain is the buoyant force, = /v , where is the rate of energy gain induced by buoyancy and = g f ( A) with f ( A) being a function of the Atwood number. The rate of momentum loss is = /v , where is the rate of dissipation of kinetic energy. Without viscous scale on the basis of dimensional grounds = C v 3 / L , where C is a constant and L is the characteristic length scale for energy dissipation, either in the vertical or horizontal direction. The model identies two distinct mechanisms of mixing development, shows that invariant, scaling and spectral properties of RT mixing depart from the canonical scenario and suggests that turbulent mixing ow has a more regular character compared to isotropic Kolmogorov turbulence [4, 6366]. Below we perform a comparative study of modeling approaches [4161]. First we show that, on the one hand, momentum consideration (3.6) can be reduced in some limiting cases to interpolation and turbulence models (3.1)(3.4), and that, on the other hand, it identies

Phys. Scr. T142 (2010) 014012

S I Abarzhi and R Rosner

new features of mixing dynamics. We then consider the mechanism and properties of turbulent mixing and discuss some macroscopic properties of the ow, such as energy budget, momentum and energy transport, diffusion and dissipation mechanisms, dynamics of the center of mass, and ow drag. We further discuss theoretical aspects of turbulent mixing dynamics, including symmetries and invariants, and analyze its scaling, spectra and uctuations. Similarities and distinctions will be identied between the non-inertial turbulent mixing and inertial Kolmogorov turbulence. Connections of the modeling results to the observations will also be discussed. 2.3.2. Similarities of the results of the models. As discussed in [5961], asymptotic solutions for model (3.6) depend on whether the characteristic length scale of the ow is horizontal or vertical. If the characteristic length scale is horizontal, L , then equations (3.6) have steady solutions with v g and h t g , and the rates of momentum and energy are balanced: = = g and = = (g )3/2 / C (hereafter for simplicity we re-scale g f ( A) g ). If the characteristic scale is vertical, L h , then asymptotically in time, h = agt 2 /2 and v = agt with a = (1 + 2C )1 . The rates of energy gain and dissipation are time dependent, = ag 2 t and = (1 a )ag 2 t , and the rates of momentum gain and loss are time invariant and scale invariant, =g and = C v 2 / h [5961]. Found in many observations, the values of a are rather small, a 0.050.15 [1739]. Thus, in the mixing ow, almost all the energy induced by the buoyancy dissipates, with / = (1 a ), and there is a slight imbalance between the rates of momentum gain and loss, with ( )/ = a . Self-similar mixing may develop when the horizontal scale grows with time as h gt 2 [43, 47] and when the vertical scale h , , is the characteristic scale for energy dissipation that hh occurs in small-scale structures at the uid interface [5961]. Results obtained within momentum consideration agree in certain approximations with the results found by interpolation and turbulent models (3.1)(3.6). For instance, the momentum model [5961] reproduces the results of Glimm and Sharp [43] if one assumes that the characteristic length scale of the ow is horizontal, L , imposes the growth h gt 2 and adjusts the values of dimensionless parameters as f ( A) = A and C = (1 2)/4 to ensure the dependence h = Agt 2 found in [43, 44]. On the other hand, if one assumes that the characteristic scale is vertical, L h , and properly re-scales the values of f ( A) and C , one can reduce the momentum consideration [5961] to the model of Youngs [45]. Further, by introducing adjustable parameters to describe the merger process and the volumetric and surface terms, one can reduce momentum model to the drag models [4749, 6163], etc. With the proper choice of the initial conditions, momentum consideration [5961] reproduces also results of turbulence models [41, 42, 5053], which consider RT mixing as an effective diffusion process. Furthermore, in agreement with results of Dimotakis [54] and Zhou et al [55, 56], momentum consideration suggests that properties of RT turbulent mixing may resemble certain properties of Kolmogorov turbulence [5961]. This situation may occur,
5

Table 1. Mechanisms of mixing development. Model [43, 44, 47, 48] [45, 46] [5961] Mechanism Mixing is induced by growth of horizontal scale gt 2 Mixing is induced by a broad-band incoherent initial perturbation Growth of horizontal scale gt 2 is possible Dominance of vertical scale h may lead to ow acceleration

for instance, when the rates of gain and loss of specic momentum and energy are balanced, = and = , and the ow is steady, so that = 0 and buoyancy forcing supplies energy at a constant rate = = (g )3/2 / C . This can account for the observations [23, 26] of statistically steady turbulence in RT ows with relatively low Reynolds numbers Re 103 . It also suggests that RT instability indeed can be applied to produce semi-isotropic Kolmogorov type of turbulent owsa well-known approach in experimental uid dynamics [54]. On the other side, in agreement with the model of Chertkov [57], the momentum consideration indicates that in accelerated turbulent mixing with h gt 2 the rate of energy dissipation grows with time as g 2 t , and, as it was done in [57], one might choose to substitute these dependencies in canonical Kolmogorov and ObukhovBolgiano scaling and spectra [6466, 70, 71]. 2.3.3. Distinctive properties of the turbulent mixing 2.3.3.1. On the mechanism of turbulent mixing evolution. Agreeing in certain limiting cases with principal results of the interpolation and turbulent diffusion models, momentum consideration identies some new properties of the mixing ow [5961]. It suggests that the accelerated turbulent mixing develops due the imbalance of gain and loss of specic momentum, = . This imbalance may occur when (i) the horizontal scale grows as gt 2 and/or when (ii) the vertical scale h is a characteristic scale for energy dissipation, = C v 3 / L with L h , and when it represents cumulative contributions of small-scale structures into the ow dynamics. Existence of two distinct mechanisms of the mixing development reconciles models [4349] and [5058]. It also agrees with results of theoretical studies [3], which and period provide independent found that the amplitude h contributions to the nonlinear RT dynamics and that for highly isotropic large-scale coherent structure the growth of horizontal scales may not occur [80], table 1. 2.3.3.2. On energy budget, transports of energy and momentum and position of the center of mass. Turbulence is a property of dissipative systems and it decays if it is not driven [15, 16, 6366]. Kolmogorov turbulence is driven by an external energy source, which supplies energy to the ow at a constant rate : energy is injected at large scales by an external source, and then it is transferred without loss through the inertial interval and dissipates at small scales [15, 16, 6366]. Interpolation and turbulent diffusion models [4156] do not consider the source of energy of turbulent mixing ow. Furthermore, turbulence models [5056] presume that the energy and momentum transports in RT mixing are similar to

Phys. Scr. T142 (2010) 014012

S I Abarzhi and R Rosner

Table 2. Energy source, transport of momentum and energy and position of the center of mass. Model [15, 16] [5057] [5961] Energy source Energy is injected at large scales by an external source and then transferred without loss through the inertial interval and dissipated at small scales. Mean velocity of the center of mass of the uid system is time independent Presume that the transport of energy and of momentum are similar to that in Kolmogorov turbulence There is no external energy source other than gravity. Energy and momentum are gained due to buoyancy and lost due to dissipation. In the steady regime = 0 and = = (g )3/2 / C . Accelerated turbulent mixing is driven by imbalance between the gain and loss of specic momentum, and at any scale = and = . In accelerated mixing, the mean velocity of the center of mass of the uid system is time dependent Table 3. Symmetries of turbulent dynamics. Model [15, 16] [5057] [5961] Symmetries Kolmogorov turbulence is inertial and is invariant with respect to Galilean transformation, translations in time and 3D space and spatial rotations and inversions. It is scale invariant, L L K , T T K 1n , v v K n with n = 1/3 Presume that symmetries of RT turbulent mixing are the same as in Kolmogorov turbulence RT turbulent mixing is non-inertial and is invariant with respect to translation, rotations and inversions in the plane normal to gravity g. It is scale invariant, L L K , T T K 1n , v v K n with n = 1/2

that in Kolmogorov turbulence. According to the momentum consideration [5961], for RT turbulent mixing an external energy source (other than gravity) is not required, and the specic momentum is gained due to buoyancy and is lost due to dissipation. In accelerated ow at any scale = and = , and this imbalance indicates that the mean velocity of the center of mass of the uid entrained in the motion is time dependent, whereas in statistically steady turbulent ow it is invariable, table 2. 2.3.3.3. On the asymptotic state in space and time. Statistically steady Kolmogorov turbulence is an asymptotic in time state, which is achieved when the memory of the initial conditions is completely lost, and when the boundaries of the outside domain do not inuence the dynamics [15, 65, 66]. These conditions can be realized in a spatially extended system or in a nite-size domain, when the span of scales runs several decades from viscous to integral scale [15]. Implementation of these conditions in RT turbulent mixing requires special attention [28]. In a nite-size domain, an asymptotic in time dynamics corresponds to a stable state with no motion at all: under the inuence of gravity (directed from the top to the bottom) the system transits from an unstable conguration to a stable conguration (e.g. from an initial state with heavier uid located at the top of the domain and lighter uidat the bottom to a reverse state), and the change in the system potential energy dissipates into heat. In a spatially extended system (e.g. in a large domain) the ow may accelerate, however at a certain time compressibility and stratication start to play a role and results in ow stabilization, as discussed in [14, 59, 73, 74]. To allow for the development of RT turbulent mixing and to enable its diagnostics over substantial span of scales, the size of the domain should be large enough yet not so large to prevent mixing stabilization by effects of compressibility and stratication. 2.3.3.4. On effective drag in the mixing ow. Regularization of accelerated turbulent mixing is at rst glance an unusual (and certainly unexpected by turbulent mixing community) concept. However, there is some evidence from previous studies that is does take place. For instance, re-laminarization
6

of an accelerated ow is a well-known uid dynamics phenomenon discovered in the works of Taylor [75] for ows in curved pipes and Narasimha and Sreenivasan [76] for boundary layers. Another indication of a more regular character of RT mixing follows from the characteristic value of the ow drag. Coefcient C in the relations = C v 3 / L and = C v 2 / L can be viewed as effective drag coefcient, which is related to the growth-rate h = agt 2 /2 via a (1 + 2C ) = 1 [5961]. For C 0 (no drag) the solution is a free-fall with a 1, whereas for C (innitely large drag) a 0 and the ow cannot accelerate. Experiments and simulations report relatively small values of a 0.050.15 (with 0.030.07 in the relation h = A gt 2 in [49]). These values correspond to drag coefcient of C 38, indicating that ow may tend to be more laminar rather than turbulent [77]. In canonical Kolmogorov turbulence, the value of C is calculated from the third-order velocity structure function as C = 5/4 and C 1 [15, 66]. This may lead to a = 2/7 0.3 ( 0.14 in h = A gt 2 in [49]), which is signicantly greater than the values actually observed [49]. 2.3.3.5. On symmetries and invariants of RT mixing. A cornerstone of Kolmogorov theory is that the isotropic and homogeneous turbulent ow has a number of symmetries in statistical sense [15, 16, 6366]. It is invariant with respect to Galilean transformation, to translations in time and in 3D space, to spatial inversions and rotations and to scaling transformation with L L K , t t K 1n and v v K n for any n , where v = |v| [15, 66]. Kolmogorov found that n = 1/3, and the measure of the scaling symmetry is the rate of change of specic kinetic energy v 3 / L . Similarly to Kolmogorov turbulence, RT turbulent mixing has a number of symmetries [4, 61], however due to the presence of gravity, g = 0, and non-inertial character of the dynamics, these symmetries are distinct from those of Kolmogorov turbulence. RT mixing ow is invariant with respect to translations, inversions and rotations in the plane normal to g, and to scaling transformation L L K , t t K 1n and v v K n with n = 1/2. The measure of this scaling symmetry v 2 / L has the same dimension as g and quanties the rate of change of specic momentum, table 3.

Phys. Scr. T142 (2010) 014012

S I Abarzhi and R Rosner

Table 4. Some invariant properties of turbulent dynamics. Model [15,16] [5057] [5961] Invariants Dynamics is statistically steady. Invariance of energy dissipation rates v 3 / L is compatible with the existence of the inertial interval and energy cascade. Enstrophy and helicity are other invariants Presume that the inertial interval and energy cascade exist in accelerated turbulent mixing and substitute the dependence L gt 2 in energy dissipation rate and other invariants of canonical turbulent ows Dynamics is statistically unsteady. Invariance of rate of momentum loss v 2 / L leads to non-dissipative specic momentum transport between the scales. Energy dissipation rate and enstrophy are time dependent. Helicity is invariant for h gt 2 Table 5. Velocity spatial scaling. Model [15, 66] [5961] Velocity scaling vl /v (l / L ) based on invariance vl /v (l / L )1/2 based on invariance
1/3

Velocity N th-order structure function (l ) N /3 based on invariance (l ) N /2 based on invariance

Table 6. Velocity temporal scaling. Model [15, 66] [5961] Velocity scaling v /v (/ T )1/3 based on invariance v /v (/ T ) based on invariance Velocity uctuations due to turbulent transport v (v )1/3 based on invariance and (v )1/3 ( )1/2 v based on invariance and ( ) g

In isotropic turbulence, the total momentum is zero because of isotropy, and time- and scale-invariance of the energy dissipation rate v 3 / L is compatible with existence of inertial interval and non-dissipative energy transfer between the scales [15, 16, 6366]. In accelerated RT turbulent mixing, the specic momentum is imbalanced, = , and time- and scale invariances of v 2 / L imply that at any time and length scale the specic momentum is being lost at the same constant rate, and momentum transfer between the scales is non-dissipative [61]. Enstrophy is another invariant of isotropic turbulence [6466], whereas in RT mixing this value decays with time. This provides another indication of a tendency of accelerated mixing ow to re-laminarize [61]. In RT ow, vortical structures form helixes not vortices. In a ow dominated by the growth of horizontal scales, h gt 2 , the helicity is a statistically steady value and its steadiness may serve as an indicator of achieving a merger-driven self-similarity [61], table 4. The interpolation models [4349] and the turbulent diffusion model [41, 42, 5057] do not discuss distinctions in symmetries and invariants of Kolmogorov turbulence and RT turbulent mixing. Models [5057] presume that similarly to Kolmogorov turbulence, inertial interval and energy cascade exist in accelerated turbulent mixing and substitutes the dependence L gt 2 in the quantities of canonical turbulent ows, table 4. 2.3.3.6. On spacetime scaling properties and the role of uctuations. For a description of scaling properties, let the length scale L and time scale T refer to large scales and times, with the characteristic velocity v . Let the characteristic velocity be vl at a small length scales l , and let the characteristics velocity be v on a short time-scale . In Kolmogorov turbulence [14, 15, 16, 6466], the invariance of the energy dissipation rate v 3 / L vl3 / l yields the velocity scaling vl /v (l / L )1/3 , N th-order velocity structure function (l ) N /3 , and velocity scaling with time v /v (/ T )1/3 . The relative velocity of two
7

parcels of uids involved in the motion is ( )1/2 on a time delay , and it is substantially smaller than the velocity uctuations v (v )1/3 induced by turbulence. This well-known result means that in Kolmogorov turbulence, the main contribution to velocity uctuations is provided by the turbulence not by the initial conditions [14, 15, 66], tables 5 and 6. In RT turbulent mixing, the invariance of the rate of momentum loss v 2 / L vl2 / l yields the velocity scaling vl /v (l / L )1/2 , N -th order velocity structure function (l ) N /2 and the velocity scaling with time v /v (/ T ). For two parcels of uids involved in the motion with a time delay , their relative velocity is ( ) g and it is comparable to v induced by the turbulent process, whereas their own velocities grow with time as gt and g (t ) [14]. We see that in accelerated mixing ow, the velocity uctuations are frozen to the level of the initial conditions, and with time the contribution of uctuations to the mixing dynamics is reduced, tables 5 and 6. 2.3.3.7. On Reynolds number, viscous scale and integral scale. In Kolmogorov turbulence, Reynolds number is nite Re = v L / = const and local Reynolds number Rel = vll / scales as Rel Re(l / L )4/3 leading to the viscous length scale l ( 3 /)1/4 and time-scale (/)1/2 for Rel 1. In accelerated turbulent mixing the Reynolds number grows with time as Re = v L / g 2 t 3 / and the local Reynolds number Rel = vl l / scales as Rel Re(l / L )3/2 . For Rel 1 viscosity plays a dominant role, thus leading to viscous length scale l ( 2 /)1/3 with the corresponding time scale (/2 )1/3 . This viscous length scale is nite. It is set by the ow acceleration and is comparable to the wavelength of mode of fastest growth [1, 2]. Thus despite in accelerated RT mixing the Reynolds number can reach large values relatively quickly the ow viscous scale remains nite. An upper limit for Reynolds number Re g 2 t 3 / can be estimated at a border of validity of incompressible approximation gt c as Rec c3 /g , where c is the sound speed; see table 7.

Phys. Scr. T142 (2010) 014012

S I Abarzhi and R Rosner

Table 7. Reynolds number, viscous scales and integral scale. Model [15, 66] [5961] Reynolds number Re = v L / = const. Invariance of leads to Rel (vl l /) Re(l / L )4/3 Re = v L / g 2 t 3 / . Invariance of leads to Rel (vl l /) Re(l / L )3/2 For gt c upper limit is Rec c3 /g Viscous and integral scales Invariance of leads to l ( 3 /)1/4 and (/)1/2 An integral scale is the scale at which energy is gained by the ow system Invariance of leads to l ( 2 /)1/3 and (/2 )1/3 An integral scale is the coarsest vertical scale representing cumulative contributions of small scale structures

Table 8. Dimensional-ground-based spectra. Model [15, 16] [5961] Specic kinetic energy E (k ) k set by invariance of E (k ) k 2 set by invariance of
2/3 5/3

Specic momentum M (k ) 0 due to isotropy M (k ) 1/2 k 3/2 set by invariance of

Table 9. Pressure uctuations. Model [15, 66] [5961]

4/3 4/3

Scaling l set by invariance of 2 l 2 set by invariance of

Spectrum k set by invariance of 2 k 3 set by invariance of

4/3 7/3

In Kolmogorov turbulence the integral scale is the scale, at which energy is gained by the ow system. For turbulent mixing this consideration may not be directly applicable. In RT mixing, momentum and energy are gained and dissipated at any scale, and imbalance between the rate of momentum gain and loss leads to ow acceleration. The coarsest vertical scale in RT ow can be regarded as an integral cumulative scale, which represents cumulative contributions of small-scale structures in the ow dynamics, table 7. 2.3.3.8. On dimensional-analysis-based spectral properties. In isotropic turbulence, the invariance of energy dissipation rate leads to kinetic energy spectrum E (k ) 2/3 k 5/3 [14, 15, 66]. In RT mixing accurate determination of spectra (and corresponding eigenfunctions) is a formidable theoretical task because the dynamics is statistically unsteady. Dimensional analysis suggests that the spectrum of the specic kinetic energy has the form E (k ) k 2 , which is steeper than Kolmogorov; similarly for the spectrum of specic momentum one obtains M (k ) 1/2 k 3/2 (in Kolmogorov turbulence M (k ) 0 due to isotropy), table 8 [61]. 2.3.3.9. On pressure uctuations. In Kolmogorov turbulence, pressure uctuations are evaluated using fourth-order velocity structure function so that pressure uctuates as 4/3l 4/3 with spectrum 4/3 k 7/3 [66]. For RT mixing dimensional grounds suggest for pressure uctuations 2l 2 with spectrum 2 k 3 , which is steeper than in Kolmogorov turbulence, table 9 [61]. 2.3.3.10. On scalar transport. An important outcome of momentum consideration [5961] is the distinct roles, which are played by diffusion and dissipation in RT mixing: diffusion inuences the gains of momentum and energy, dissipation affects their losses and the scalar transport departs from a standard form of Fickian diffusion. Based on energy conservation (per unit volume), the heat transport is described by the equation ( s )/ t ( /)2 [ ] = 0, (4.1)
8

where is temperature, s is entropy, is thermal conductivity and [ ] denote other terms [14, 64, 65]. In the Boussinesq approximation, one presumes that the change in density induced by temperature contrast is substantially greater than the change in density induced by change in hydrostatic pressure (e.g. by gravitational stratication), and that both are smaller than the uid density itself [14]. In isentropic limit energy, equation (4.1) is then reduced to diffusion equation of Fickian type: / t 2 + (v ) = 0, (4.2)

where is thermal diffusivity. Temperature uctuations are described by thermal dissipation function = ( )2 . For turbulent ow v L and (v/ L )( )2 [14, 65]. The passive scalar consideration (4.2) works remarkably well for canonical problems of isotropic and homogeneous turbulence [65] and turbulent convection [71, 72]. In these ows, the thermal dissipation function is timeand scale invariant, similarly to energy dissipation rate in Kolmogorov turbulence. In the case of isotropic and homogeneous turbulence, the invariance of (v/ L ) 2 (vl / l )l2 results in temperature uctuations with l2 1/3l 2/3 . In the case of turbulent convection, accounting for the constancy of temperature contrast, temperature uctuates 4/5 as l2 (g)2/5l 2/5 , where is a thermal expansion coefcient [64, 65, 71, 72]. Momentum consideration [5961] suggests that thermal transport inuences the rate of momentum gain = g / and that in the isentropic limit / / the equation (4.1) yields for L h asymptotically in time (v/gL ) 2, (t )[h /gt 2 , v/gt , /, /g , / g ] 1,

(4.3) where (t ) = ln(gt / h 0 ) and h 0 is an initial length scale. In contrast to problems of isotropic turbulence and turbulent convection [64, 65, 71, 72], in RT mixing thermal dissipation function is not an invariant, and the rates of momentum and energy , , , , are time dependent. This limits
2

2 (t )[/g 2 t , /g 2 t , t / 2 ] 1,

Phys. Scr. T142 (2010) 014012

S I Abarzhi and R Rosner

Table 10. Thermal dissipation and temperature uctuations. Model [64, 65, 71, 72] [5961] Thermal dissipation In isotropic turbulence and convection thermal dissipation function = ( )2 is an invariant In RT mixing values , , , , are time-dependent and value = / = / 1 is invariant Temperature uctuations In isotropic turbulence l2 1/3 l 2/3 ; 4/ 5 in convection l2 (g)2/5 l 2/5 2 Constancy of v /(/)gL can be used in sub-grid-scale models to account for scalar transport and contributions of small scales

Table 11. Dissipative scale. Model Batchelor scale [64, 78] lB = (B )1/2 based on invariance of and lB = l Sc1/2 [5961] lB ( 2/2 )1/6 based on invariance of and lB = l Sc1/2

ones capabilities to rigorously estimate scaling and spectral properties of the mixing ow [61]. However, given that (t ) = ln(gt 2 / h 0 ) is a slowly growing function, at large but nite times the viscous and dissipative scales of the ow would still remain nite. It is remarkable that in the mixing ow the ratio = / = / is a constant value with 1 at highly nonlinear and turbulent stages, with or without turbulent diffusion accounted for. Parameter v 2 /(/ )gh can be thus be used as the ow characteristics in the case of sustained and/or time-dependent accelerations (for instance, for evaluation of the properties of Reynolds stress in the sub-grid-scale models), table 10. We point the readers attention to the fact that the momentum consideration [5961] reproduces the results of turbulent diffusion models [41, 42, 5056] when one neglects the effect of scalar transports on the gains of momentum and energy. This case can be realized in experiments when the scalar is a neutral buoyancy dye with zero heat solubility. Then, employing expression for viscous scale l ( 2 /)1/3 and considering the dynamics within the context of passive scalar mixing [64, 78], one can nd for RT ows the dissipative (Batchelor) length scale lB ( 2 /2 )1/6 . Similarly to l , this scale is nite and set by ow acceleration. For Sc 1, where Sc = (/) is the Schmidt number, lB l , table 11. 2.3.3.11. On 2D and 3D ows. The 2D and 3D dynamics often exhibit drastically different properties, and turbulent ows are no exceptions. For instance, in the case of isotropic and homogeneous turbulence, 3D ow exhibits the direct cascade, whereas 2D ow exhibits the inverse cascade [15, 6366, 79]. Extending these classical results to the case of RT dynamics via the use of Boussinesq approximation and substitution of time dependence L gt 2 in the energy dissipation rate v 3 / L g 2 t and in the thermal dissipation function (v/ L ) 2 2 / t , model [57] claimed that 3D RT mixing is driven by energy transport, and similarly to isotropic turbulence, it is characterized by velocity uctuations with vl /v (l / L )1/3 and temperature uctuations with l / (l / L )1/3 . This leads to viscous scale l ( 3 /)1/4 ( 3 /g 2 t )1/4 and dissipative scale lB ( 3 /g 2 t )1/4 Sc1/2 . For the 2D RT ow, model [57] established that the inverse cascade cannot be realized because the pumping scale L gt 2 grows too fast for the larger structures to become established and
9

suggested applying the scenario of turbulent convection. Based on the time-dependent thermal dissipation function (v/ L ) 2 2 / t , model [57] derives the scaling vl /v (l / L )3/5 and l / (l / L )1/5 and identies viscous and dissipative scales as l (/v)5/8 L 3/8 5/8 t 1/8 /g 1/4 and lB Sc1/2 ( 5/8 t 1/8 /g 1/4 ). Momentum consideration [5961] nds that RT ow is driven by momentum transport, and its viscous and dissipative scales are nite and set by ow acceleration, in either 2D or 3D ow (table 12). The other turbulent and interpolation models [43, 44, 4749] do not distinguish between the dynamics of 2D and 3D ows and suggest the growth of horizontal scale gt 2 , induced by binary interactions, as the sole mechanism of mixing development. Group theory consideration [80] conrms their results only partially and notes that, in a 3D ow, the interactions are essentially multi-pole and that the growth of horizontal scale may not occur if the large-scale coherent dynamics is highly isotropic. The latter result agrees with momentum consideration [5961], which does not require the growth of horizontal scales for mixing evolution (table 12). 2.3.3.12. On statistically steady and statistically unsteady dynamics. To conclude this section, we discuss in more detail statistically steady and unsteady regimes in RT ows. In a steady regime, the ow can appear more coherent or more turbulent depending on the Atwood number and the initial conditions [3, 4]. For the steady ow, the rates of momentum gain and loss as well as energy gain and dissipation are balanced, = and = , and the characteristic length scale of the ow is constant. The characteristic velocity is v g , the Reynolds number is Re = v L / g / and the energy dissipation rate is constant (g )3/2 /. This formally corresponds to the viscous scale ( 3 /)1/4 ( 3 /(g )3/2 )1/4 , which is smaller than the mode of fastest growth ( 2 /g )1/3 for > ( 2 /g )1/3 . However, as /( 3 /(g )3/2 )1/4 (/( 2 /g )1/3 )9/8 and 9/8 = 1.125, the characteristic span of scales in the steady ow is well captured by the ratio /( 2 /g )1/3 . The ow acceleration increases the ow velocity, integral length scale, Reynolds numbers and energy dissipation rate. At rst glance, this may lead to the appearance of high Reynolds number turbulent ow with a signicant span of scales [4158]. Momentum consideration [5961] suggests, however, that buoyancy-driven turbulent mixing is accelerated due to an imbalance between the gain and loss of momentum and energy with ( )/ = a and ( )/ = a . In this ow the velocity v gt , the length scale h gt 2 , the Reynolds number Re g 2 t 3 / and the span of scales gt 2 /( 2 /g )1/3 indeed increase. Here the viscous scale

Phys. Scr. T142 (2010) 014012

S I Abarzhi and R Rosner

Table 12. 2D and 3D dynamics. Model [57] 3D ow 2D ow Substitute L gt 2 in (v/ L ) 2 suggest scaling vl /v (l / L )3/5 and l / (l / L )1/5 as in turbulent convection and nds for viscous and dissipative scales l 5/8 t 1/8 /g 1/4 and lB Sc1/2 ( 5/8 t 1/8 /g 1/4 ) Presume that growth gt 2 is driven by binary interactions and it is the sole mechanism of the mixing development Find that growth of horizontal scales is driven by binary interactions Compared to 3D ows, regularization process may be harder to implement in 2D ows Substitute L gt 2 in v 3 / L suggest scaling vl /v (l / L )1/3 and l / (l / L )1/3 as in isotropic turbulence and nd for viscous and dissipative l ( 3 /g 2 t )1/4 and lB ( 3 /g 2 t )1/4 Sc1/2 [43, 44] [4749] Presume that growth gt 2 is driven by binary interactions and it is the sole mechanism of the mixing development [80] Find that growth of horizontal scales is driven by multi-pole interactions and may not occur for highly isotropic coherent structures [5961] Find that mixing may develop due to the dominance of vertical scale. Find that viscous and dissipative scales are nite and set by ow acceleration. Suggest that RT mixing may regularize with proper choice of the initial conditions and acceleration history

Table 13. Steady and accelerated RT ows. [5961] Steady ow Flow quantities Balance of momentum and energy = and = . Constant length scale , velocity v g , 3/2 Reynolds number Re g / and energy dissipation rate (g ) / with corresponding viscous scale ( 3 /(g )3/2 )1/4 and span of scales (/( 2 /g )1/3 )9/8 Accelerated ow Imbalance of momentum and energy ( )/ = a and ( )/ = a . Time-dependent length scale h gt 2 , velocity v gt , Reynolds number Re g 2 t 3 / and energy dissipation rate g 2 t . Constant rate of momentum loss g with corresponding viscous scale ( 2 /g )1/3 and span of scales gt 2 /( 2 /g )1/3 (upper limit c2 /g ( 2 /g )1/3 )

is ( 2 /g )1/3 and the upper limit for the span of scales is (c/(g )1/3 )2 for c gt . However, compared to the case of statistically steady isotropic and homogeneous turbulence, accelerated turbulent mixing exhibits stronger correlations, reduced contribution of uctuations and steeper spectra, and tends to be more laminar [5961, 76, 77]; see table 13. 2.4. Connection to observations As discussed earlier, momentum consideration [5961] agrees in some limiting cases with the results of interpolation and turbulent diffusion models [4158]. It can thus be applied to interpret the existing observations [1739] to the same extent as the models [4158]. Furthermore, the momentum consideration [5961] is self-consistent and explains the observations, which other models do not explicitly address. For instance, a relatively ordered character of RT mixing identied within momentum consideration agrees with the experiments [56], which found that at very high Reynolds numbers Re > 105 the mixing ow keeps a signicant degree of order. On the other hand, momentum consideration suggests that at relatively low Reynolds number Re 103 , RT mixing may resemble some of the properties of statistically steady turbulence (conditional on whether there is a balance between the gains and losses of specic momentum and energy, as well as on the density ratio and initial perturbation). Departures from the canonical Kolmogorov scenario, which were found in recent experiments and simulations [1739], indicate the importance of momentum transport for the ow dynamics, in agreement with [5961]. The frozen character of uctuations in RT mixing indicates that this ow is more sensitive to the initial conditions compared to the case of canonical turbulent ows, in agreement with observations [1739]. Stronger correlations and coupling and reduced level of uctuations are indications of a higher degree of order in RT
10

mixing compared to the case of isotropic and homogeneous turbulence [5961]. This principal theoretical result opens up new opportunities for the design of experiments and simulations, which may include realization of either a more turbulent and strongly uctuating mixing ow or a more regular and better controlled mixing ow at high Reynolds numbers [1758]. Furthermore, as suggested by the relative independence of horizontal and vertical scales in the mixing ow, these opportunities can be realized in a broad parameter regime. For instance, horizontal scales can be controlled with the proper choice of the dispersion relation, initial conditions, isotropy and coherence, whereas vertical scales can be inuenced by the means of time-dependent and spatially varying acceleration and deceleration, diffusion and stratication, etc [3, 4, 5961]. Details of the potential design of such experiments and simulations will be provided in a forthcoming publication. The momentum consideration [5961] indicates that for a reliable description of RT mixing it is essential to account for the interface dynamics (1) and (2), and that the analysis of 4D momentumenergy transport describing compressible turbulent mixing may help us to cut through the problems complexity. Our comparative study of different approaches for modeling RT turbulent mixing [4161] offers to experiments and simulations a number of diagnostic parameters and invariant measures, including the position of the center of mass, energy budget, momentum transport, scalar transport, effective drag, scaling and spectra, correlations and uctuations, etc. Based on these results, a number of qualitative experimental tests can be performed (e.g. ow regularization at high Reynolds number and distinct mechanisms of mixing development), and some robust quantitative parameters can be implemented in numerical simulations to model sub-grid scales and to capture the properties of the Reynolds stress

Phys. Scr. T142 (2010) 014012

S I Abarzhi and R Rosner

(e.g. the value of = / = / 1 at any scale, for sustained and time-dependent acceleration, with and without accounting for turbulent diffusion, in either steady or unsteady regimes [5961]). Direct testing of the momentum consideration [5961] would require further enhancements in ow diagnostics and in the quality and information capacity of experimental and numerical data sets, including tight control of macroscopic parameters, accuracy, precision, acquisition rates and dynamic range [28]. It therefore has a potential to bring experiments and simulations to a new level of standard.

3. Conclusions
In this work, we have performed a comparative study of different approaches [4161] for modeling RT turbulent mixing. Interpolation models, turbulent diffusion models and momentum consideration [4161] agree with one another under certain approximations and can be applied for interpreting existing experimental and numerical data sets [1739]. By focusing on momentum transport [5961], one can identify a number of new properties of RT ows. These include, among others, two distinct mechanisms of mixing development, non-dissipative transport of specic momentum between the scales, time- and scale-independence, the ratio between the rates of loss and gain of specic momentum and energy = / = / for sustained and varying acceleration, and nite values of viscous and dissipative scales. Compared to isotropic, homogeneous and statistically steady turbulence, the anisotropic, inhomogeneous and statistically unsteady RT turbulent mixing is characterized by a different set of symmetries and invariants. It has stronger correlations and coupling, weaker contribution of uctuations to the ow dynamics, stronger dependence on the initial conditions, and steeper dimensional-analysis-based spectra. The more ordered character of the mixing dynamics opens up new opportunities for the design of experiments and simulations for the realization of regular or turbulent high Reynolds number mixing ows.

Acknowledgments
SIA thanks the US National Science Foundation and the Ofce of Fusion Energy Sciences of the US Department of Energy for support, Dr K R Sreenivasan and S Gauthier for deep discussions and Drs S I Anisimov, B Alder, A Bershadskii, R P Drake, S Fedotov, B Fryxell, Y Fukumoto, G Glatzmaier, G Hazak, L P Kadanoff, A Klimenko, V Lvov, K Nishihara, A Pouquet, B Remington, J Schumacher, A L Velikovich, J Werne and V Yakhot for discussions and comments.

References
[1] Rayleigh L 1883 Investigations of the character of the equilibrium of an incompressible heavy uid of variable density Proc. London Math. Soc. 14 170 [2] Davies R M and Taylor G I 1950 The mechanics of large bubbles rising through extended liquids and through liquids in tubes Proc. R. Soc. A 200 375 11

[3] Abarzhi S I 2008 Review on nonlinear coherent dynamics of unstable uid interface: conservation laws and group theory Phys. Scr. T132 297681 [4] Abarzhi S I 2010 Review of theoretical modeling approaches of RayleighTaylor instabilities and turbulent mixing Phil. Trans. R. Soc. A 368 1809 [5] Zeldovich Ya B and Raizer Yu P 2002 Physics of Shock Waves and High-Temperature Hydrodynamic Phenomena 2nd edn (New York: Dover) [6] Remington B A, Drake R P and Ryutov D D 2006 Experimental astrophysics with high power lasers and Z -pinches Rev. Mod. Phys. 78 755 [7] Drake R P 2009 Perspectives of high energy density physics Phys. Plasmas 16 055501 [8] Hillebrandt W and Niemeyer J C 2000 Type Ia supernova explosion models Annu. Rev. Astron. Astrophys. 38 191 [9] Spalart P R and Watmuff J H 1993 Experimental and numerical study of a turbulent boundary layer with pressure gradients J. Fluid Mech. 249 337 [10] Gutman E J, Schadow K C and Yu K H 1995 Mixing enhancement in supersonic free shear ows Annu. Rev. Fluids 27 375 [11] Stone J M, Hawley J F, Gammie C F and Balbus S A 1996 Three-dimensional magneto-hydrodynamical simulations of vertically stratied accretion disks ApJ 463 656 [12] Choi J P and Chan V W S 2002 Predicting and adapting satellite channels with weather-induced impairments IEEE Trans. Aerosp. Electron. Syst. 38 779 [13] Marmottant P and Villermaux E 2004 On spray formation J. Fluid Mech. 498 73 [14] Landau L D and Lifshitz E M 1987 Course of Theoretical Physics VI: Fluid Mechanics (New York: Pergamon) [15] Kolmogorov A N 1941 Local structure of turbulence in an incompressible uid for very large Reynolds numbers Dokl. Akad. Nauk SSSR 30 299 Kolmogorov A N 1941 Energy dissipation in locally isotropic turbulence Dokl. Akad. Nauk SSSR 32 19 [16] Sreenivasan K R 1999 Fluid turbulence Rev. Mod. Phys. 71 S383 [17] Read K I 1984 Experimental investigation of turbulent mixing by RayleighTaylor instability Physica D 12 45 [18] Marinak M M, Glendinning S G, Wallace R J, Remington B A, Budil K S, Haan S W, Tipton R E and Kilkenny J D 1998 Nonlinear RayleighTaylor evolution of a three-dimensional multimode perturbation Phys. Rev. Lett. 80 4426 [19] Schneider M, Dimonte G and Remington B 1998 Large and smalls scale structures in RayleighTaylor mixing Phys. Rev. Lett. 80 3507 [20] Dalziel S B, Linden P F and Youngs D L 1999 Self-similarity and internal structure of turbulence induced by RayleighTaylor instability J. Fluid Mech. 399 1 [21] Dalziel S B, Patterson M D, Cauleld C P and Coomaraswamy I A 2008 Mixing efciency in high-aspect-ratio RayleighTaylor experiments Phys. Fluids 20 065106 [22] Waddell J T, Jacobs J W and Niederhaus C E 2001 Experimental study of RayleighTaylor instability: low Atwood number liquid systems with single-mode initial perturbations Phys. Fluids 13 1263 [23] Wilson M and Andrews M J 2002 Spectral measurements of RayleighTaylor mixing at small Atwood number Phys. Fluids 14 938 [24] Kucherenko Y A, Shestachenko O E, Balabin S I and Pylaev A P 2003 RFNC-VNIITF multi-functional shock tube for investigating the evolution of instabilities in non-stationary gas dynamic ows Laser Part. Beams 21 381 [25] Kucherenko Y A, Balabin S I, Ardashova R I, Kozelkov O E, Dulov A V and Romanov I A 2003 Experimental study of the inuence of the stabilizing properties of transitional layers on the turbulent mixing evolution Laser Part. Beams 21 369

Phys. Scr. T142 (2010) 014012

S I Abarzhi and R Rosner

[26] Ramaprabhu P and Andrews M J 2003 Experimental investigation of RayleighTaylor mixing at small Atwood numbers J. Fluid Mech. 503 233 Banerjee A, Kraft W N and Andrews M J 2010 Detailed measurements of a statistically steady RayleighTaylor mixing layer from small to high Atwood numbers J. Fluid Mech. 659 127 [27] Meshkov E E 2006 Studies of Hydrodynamic Instabilities in Laboratory Experiments Sarov, FGYC-VNIIEF, ISBN 5-9515-0069-9 (in Russian) [28] Orlov S S, Abarzhi S I, Oh S B, Barbastathis G and Sreenivasan K R 2010 High-performance holographic technologies for uid dynamic experiments Phil. Trans. R. Soc. A 368 1705 [29] Linden P F, Redondo J M and Youngs D L 1994 Molecular mixing in RayleighTaylor instability J. Fluid Mech. 265 97 [30] He X Y, Zhang R Y, Chen S Y and Doolen G D 1999 On the three-dimensional RayleighTaylor instability Phys. Fluids 11 1143 [31] Cook A W and Dimotakis P E 2001 Transition stages of RayleighTaylor instability between miscible uids J. Fluid Mech. 443 69 [32] Calder A C et al 2002 On validating an astrophysical simulation code ApJS 143 201 [33] Poujade O 2002 RayleighTaylor turbulence is nothing like Kolmogorov turbulence in the self-similar regime Phys. Rev. Lett. 97 085002 [34] Dimonte G et al 2004 A comparative study of the turbulent RayleighTaylor instability using high-resolution three-dimensional numerical simulations: the Alpha-Group collaboration Phys. Fluids 16 1668 [35] Cook A W, Cabot W and Miller P 2004 The mixing transition in RayleighTaylor instability J. Fluid Mech. 511 333 [36] Ristorcelli J R and Clark T T 2004 RayleighTaylor turbulence: self-similar analysis and direct numerical simulations J. Fluid Mech. 507 213 [37] Milovich J L, Amendt P, Marinak M and Robey H 2004 Multi-mode short wavelength perturbation growth studies for the National Ignition Facility double-shell ignition target design Phys. Plasmas 11 1552 [38] Cabot W H and Cook A W 2006 Reynolds number effects on RayleighTaylor instability with possible implications for type-Ia supernovae Nat. Phys. 2 562 [39] Kadau K, Rosenblatt C, Barber J L, Germann T C, Huang Z B, Carles P and Alder B J 2007 The importance of uctuations in uid mixing Proc. Natl Acad. Sci. USA 104 774107745 Kadau K, Barber J L, Germann T C, Holian B L and Alder B J 2010 Atomistic methods in uid simulation Phil. Trans. R. Soc. 368 1547 [40] Abarzhi S I and Sreenivasan K R 2010 Introduction turbulent mixing and beyond Phil. Trans. R. Soc. A 368 153946 [41] Belenki S Z and Fradkin E S 1965 Theory of turbulent mixing Trudi FIAN 29 207 (in Russian) [42] Neuvazhaev V E 1975 Theory of turbulent mixing Dokl. Akad. Nauk SSSR 222 1053 [43] Sharp D H 1984 An overview of RayleighTaylor instability Physica D 12 3 Glimm J and Sharp D H 1990 Chaotic mixing as a renormalization-group xed-point Phys. Rev. Lett. 64 2137 [44] George E, Glimm J, Li X-L, Marchese A and Xu Z-L 2002 A comparison of experimental theoretical, and numerical simulation RayleighTaylor mixing rates Proc. Natl Acad. Sci. USA 99 2587 [45] Youngs D L 1984 Numerical simulations of turbulent mixing by RayleighTaylor instability Physica D 12 32 [46] Youngs D L 1991 Three-dimensional numerical simulations of turbulent mixing by RayleighTaylor instability Phys. Fluids A 3 1312 [47] Alon U, Hecht J, Mukamel D and Shvarts D 1994 Scale-invariant mixing rate of hydrodynamically unstable interfaces Phys. Rev. Lett. 72 2867 12

[48] Alon U, Hecht J, Offer D and Shvarts D 1995 Power-laws and similarity of RayleighTaylor and RichtmyerMeshkov mixing fronts at all density ratios Phys. Rev. Lett. 74 534 [49] Dimonte G 2000 Spanwise homogeneous buoyancy-drag model for RayleighTaylor mixing and experimental evaluation Phys. Plasmas 7 2255 [50] Besnard D C, Harlow F H, Rauenzahn R M and Zemach C 1996 Spectral transport model for turbulence Theor. Comput. Fluid Dyn. 8 1 [51] Steinkamp M J, Clark T T and Harlow F H 1999 Two-point description of two-uid turbulent mixingI. Model formulation Int. J. Multiph. Flow 25 599 Steinkamp M J, Clark T T and Harlow F H 1999 Two-point description of two-uid turbulent mixingII. Numerical solutions and comparisons with experiments Int. J. Multiph. Flow 25 639 [52] Clark T T 2003 A numerical study of the statistics of a 2D RayleighTaylor mixing layer Phys. Fluids 15 2413 [53] Ristorcelli J R and Clark T T 2004 RayleighTaylor turbulence: self-similar analysis and direct numerical simulations J. Fluid Mech. 507 213 [54] Dimotakis P E 2000 The mixing transition in turbulent ows J. Fluid Mech. 409 69 [55] Zhou Y 2001 A scaling analysis of turbulent ows driven by RayleighTaylor and RichtmyerMeshkov instabilities Phys. Fluids 13 538 [56] Zhou Y et al 2003 Progress in understanding turbulent mixing induced by RayleighTaylor and RichtmyerMeshkov instabilities Phys. Plasmas 10 1883 Robey H F, Zhou Y, Buckingham A C, Keiter P, Remington B A and Drake R P 2003 The time scale for the transition to turbulence in a high Reynolds number, accelerated ow Phys. Plasmas 10 614 [57] Chertkov M 2003 Phenomenology of RayleighTaylor turbulence Phys. Rev. Lett. 91 115001 [58] Andronov V A, Bakhrakh S M, Meshkov E E, Mokhov V N, Nikiforov V V, Pevnitskii A V and Tolshmyakov A I 1976 Turbulent mixing at contact surface accelerated by shock waves Sov. Phys.-JETP 44 424 Andronov V A, Bakhrakh S M, Meshkov E E, Nikiforov V V, Pevnitskii A V and Tolshmyakov A I 1982 An experimental investigation and numerical modeling of turbulent mixing in one-dimensional ows Sov. Phys. Dokl. 27 393 Gauthier S and Bonnet M 1990 A k model for turbulent mixing in shocktube ows induced by RayleighTaylor instability Phys. Fluids A 2 1685 Soufand D, Gregoire O, Gauthier S and Schiestel R 2002 A two-time-scale model for turbulent mixing ows induced by RayleighTaylor and RichtmyerMeshkov instabilities Flow, Turbulence and Combust. 69 123 Egoire O, Soufand D and Gauthier S 2005 A second-order turbulence model for gaseous mixtures induced by RichtmyerMeshkov instability J. Turbulence 6 1 [59] Abarzhi S I, Gorobets A and Sreenivasan K R 2005 Turbulent mixing in immiscible, miscible and stratied media Phys. Fluids 17 081705 [60] Abarzhi S I, Cadjun M and Fedotov S 2007 Stochastic model of RayleighTaylor turbulent mixing Phys. Lett. A 371 457 [61] Abarzhi S I 2010 On fundamentals of RayleighTaylor turbulent mixing Europhys. Lett. 91 35000 [62] Fermi E and von Newman J 1951 Taylor instability of an incompressible liquid 26 E Fermi 1962 Collected Papers vol 2 (Chicago, IL: The University of Chicago Press) p 816 [63] Taylor G I 1935 Statistical theory of turbulence Proc. R. Soc. 151 421 [64] Batchelor G K 1953 The Theory of Homogeneous Turbulence (Cambridge: Cambridge University Press) [65] Monin A S and Yaglom A M 1975 Statistical Fluid Mechanics vol 2 (Cambridge, MA: MIT Press) [66] Frisch U 1995 Turbulence: The Legacy of AN Kolmogorov (Cambridge: Cambridge University Press)

Phys. Scr. T142 (2010) 014012

S I Abarzhi and R Rosner

[67] Gauthier S and Creurer B 2010 Compressibility effects in RayleighTaylor instability induced ows Phil. Trans. R. Soc. A 368 1618 [68] Layzer D 1955 On the instability of superposed uids in a gravitational eld ApJ 122 1 [69] Birkhoff G and Carter D 1957 Rising plane bubbles J. Math. Mech. 6 769 [70] Garabedian P R 1957 On steady-state bubbles generated by Taylor instability Proc. R. Soc. A 241 423 [71] Bolgiano J R 1959 Turbulent spectra in a stably stratied atmosphere J. Geophys. Res. 64 2226 [72] Yakhot V 1992 4/5 Kolmogorov law for statistically stationary turbulenceapplication to high-Rayleigh number Bnard convection Phys. Rev. Lett. 69 769 [73] Boffetta G, Mazzino A, Musacchio S and Vozella L 2009 Kolmogorov scaling and intermittency in RayleighTaylor turbulence Phys. Rev. E 79 065301

[74] Scagliarini A, Biferale L, Toschi F, Sbragaglia M and Sugiyama K 2010 Numerical simulations of compressible RayleighTaylor turbulence in stratied uids Phys. Scr. T142 014017 [75] Taylor G I 1929 The criterion for turbulence in curved pipes Proc. R. Soc. A 124 243 [76] Narasimha R and Sreenivasan K R 1973 Relaminarization in highly accelerated turbulent boundary layers J. Fluid Mech. 61 417 [77] Tennekes H and Lumley J L 1972 A First Course in Turbulence (Cambridge, MA: MIT Press) [78] Shraiman B I and Siggia E D 2000 Scalar turbulence Nature 405 639 [79] Kraichnan R H and Mongomery D 1980 Two-dimensional turbulence Rep. Prog. Phys. 43 647 [80] Abarzhi S I 2008 Coherent structures and pattern formation in RayleighTaylor turbulent mixing Phys. Scr. 78 015401

13