Dynamics of the Coarse-Grained Vorticity

Transcription

1 (B) Dynamics of the Coarse-Grained Vorticity See T & L, Section 3.3 Since our interest is mainly in inertial-range turbulence dynamics, we shall consider first the equations of the coarse-grained vorticity. However, it pays to consider a somewhat more general problem of the dynamics of coarse-grained velocity-gradients. Taking as the starting point the equation t ū l +(ū l )ū l = f s l p l + ν ū l + f l for the coarse-grained velocity, one easily derives, by taking the gradient with respect to x, that where D t A l = A 2 l + H l + F s l + ν A l + F l A ij u i = velocity-gradient tensor x j H ij = 2 p x i x j = pressure-tension D l,t = t + ū l x = material derivative of coarse-grained velocity (F s l ) ij = f s li x j = subscale force-gradient F ij = f i x j = external force-gradient (7) A further simplification can be made if one uses the trace-free condition, tr(a l ) = 0, which implies that tr(a 2 l) + tr(h l ) + tr(f s l ) + tr(f l)=0 Multiplying by 1/d and subtracting from the previous equation gives ( D t A l = A 2 l 1 ) d tr(a2 l)i + H l + F s l + ν A l + F l (8) where the notation M for a d d square matrix M denotes in general the trace-free or deviatoric part of M, i.e. 12

2 M = M 1 d tr(m)i There is a considerable history of phenomenological closure modelling of equation (8), going back to the seminal work of P. Vieillefosse, Local interaction between vorticity and shear in a perfect incompressible flow, J. Physique (Paris) (1982); Internal motion of a small element of fluid in an inviscid flow, Physica A (1984) who introduced the simplest approximation of setting Hl = F s l = ν A l = Fl = 0, the so-called Vieillefosse model or restricted Euler model D t A l = (A 2 l 1 d tr(a2 l)i). This model has been much studied subsequently, notably by B. J. Cantwell, Exact solution of a restricted Euler equation for the velocity gradient, Phys. Fluids A A (1992) who obtained an exact solution in terms of elliptic functions. Many further refinements of this approach, with models for ν A l, H l and (for l 0) F s l have been proposed. See C. Meneveau, Lagrangian dynamics and models of the velocity gradient tensor in turbulent flows, Annu. Rev. Fluid Mech (2011) for recent results and a rather complete bibliography of the literature. One of the interesting results of statistical models of this type is that they naturally predict that ω l S l ω l > 0 and more refined statistical alignment trends of turbulent velocity gradients, such as the preferential alignment of ω l with the intermediate strain direction, as observed in DNS by W. Ashurst, A. Kerstein, R. Kerr and C. Gikson, Alignment of vorticity and scalar gradient with strain rate in simulated Navier-Stokes turbulence, Phys. Fluids (1987) 13

3 Thus it seems that we have an explanation of the fundamental observation that ω l S l ω l > 0! The difficulty is that all of these models make fairly crude and hand-waving approximations of the unclosed terms, in particular the pressure-hessian H l. While the models seem to make remarkably good predictions for the statistics of A l, it is not clear that the modelling approximations yield accurate enough results for H l itself (and other unclosed terms). The good news is that these models suggest that much of the observed features of A l can be understood from the local self-stretching dynamics contained in the closed term A 2 l. The equation for the velocity-gradient can be separated into equations for the symmetric part D t S l = [S 2 l + Ω 2 l 1 d tr(s2 l + Ω 2 l)] + H l ( F s l and anti-symmetric part D t Ω l = (S l Ω l + Ω l S l )+ 1 2 ( F s l + F st l )+ 1 2 (F l + F l )+ν Sl. F st l )+ 1 2 (F l F l )+ν Ωl. Although these equations are superficially similar, there are profound differences. For example, note that the (nasty) pressure-hessian term is absent from the equation for Ω l. In fact, the latter equation is mathematically equivalent to the equation for ū l! It is more commonly written in terms of the filtered vorticity vector ω l, using which gives or Ω ij = 1 2 ɛ ijkω k, ω i = ɛ ijk Ω jk, D t ω l = S l ω l + (f s l + f l )+ν ω l, or ( t + ū l ) ω l =( ω l )ū l + (f s l + f l )+ν ω l (using Ω l ω l = 0!) (9) t ω l = (ū l ω l + f s l + f l ν ω l ) (using vector calculus identities) (10) 14

4 An important feature of this equation is that it has the general conservation form t ( ω l ) i + j (Σ l ) ji = 0 (11) where (Σ l ) ji = space flux of the ith-component of vorticity in the jth direction and is an anti-symmetric matric (Σ l ) ij = ū li ω lj ū lj ω li advection of vorticity stretching of vorticity + ν( ω li ω lj ) x j x i viscous diffusion of vorticity + ɛ ijk ( fl s transport of vorticity by turbulent subscale force + fl ) k (12) transport of vorticity by external force Note that we can replace ɛ ijk f s lk with the contribution from the turbulent vortex force ɛ ijk f v lk = τ l(u i,ω j ) τ l (u j,ω i ) since the difference, ɛ ijk (f s lk f v lk )= ɛ ijk k k l, does not contribute to the divergence j (Σ l ) ji. Thus, we see that the turbulent term comes from subscale vorticity advection and stretching. Although (11) has the form of a conservation law, it is not a rather unconventional conservation law since, generally, V ωd3 x = 0! Indeed, using j (x i ω j )=δ ij ω j + x i ( j ω j ) = ω i, one can see that =0 V ωd3 x = v x(ω ˆn)dA = 0, if ω ˆn = 0 at the surface V. This will be true if ω is a compactly supported vorticity distribution away from any boundaries. It will also be true in the presense of boundaries, since stick boundary conditions on the velocity (u = 0) imply that ω ˆn = 0, i.e. the normal component of ω at the boundary vanishes. Thus, the conserved quantity V ωd3 x is trivial! 15

5 A nontrivial quantity is the enstrophy Ω= 1 2 V ω 2 d 3 x. The local balance equation for largescale enstrophy can be obtained from (10), as: t ( 1 2 ω l 2 )+ [ large-scale advection {}}{ 1 2 ω l 2 ū l + viscous diffusion turbulent diffusion {}}{ ω l fl s ν ( 1 2 ω l 2 ) ] = ω l S l ω l production of large-scale enstrophy by vortex-stretching + ( ω l ) f s l flux of large-scale enstrophy to the small-scales ν ω l 2 viscous dissipation of large-scale enstrophy + ω l ( f l ) production of large-scale enstrophy by external force (13) This does not have the conservation form even formally in the limit as ν 0 and l 0, because of the production by vortex-stretching. This balance equation can be written in other, equivalent forms. For example, replacing f s l f v l gives t ( 1 2 ω l 2 )+ k [ 1 2 ω l 2 ū lk + ω li (τ l (ω i,u k ) τ l (ω k,u i )) ν k ( 1 2 ω l 2 )] = ω l S l ω l + ω lj,k (τ l (ω j,u k ) τ l (ω k,u j )) ν ω l 2 (14) Using the relations this can also be written as j τ l (ω j,u i )=τ l (ω j,u i,j )=τ l (ω j,s ij ) t ( 1 2 ω l 2 )+ k [ 1 2 ω l 2 ū lk + ω li τ l (ω i,u k ) ν k ( 1 2 ω l 2 )] = ω l S l ω l + ω li τ l (ω j,s ij ) + ω li,j τ l (ω i,u j ) ν ω l 2 (15) This is the form considered by T & L, eq. (3.3.36). A corresponding equation can be written for the small-scale enstrophy ζ l = 1 2 τ l(ω i,ω i ) which takes the form 16

6 large-scale advection [ {}}{ t ζ l + k ζ l ū lk + small-scale advection {}}{ 1 2 τ l(ω i,ω i,u k ) + turbulent diffusion viscous diffusion {}}{ ] ω li τ l (ω k,u i ) ν k ζ l = τ l (S ij,ω i,ω j )+τ l (ω i,ω j )S ij +2τ l (ω i,s ij ) ω j vortex stretching ω li,k (τ l (ω i,u k ) τ l (ω k,u i )) flux of enstrophy from large-scales ντ l (ω i,k,ω i,k ) viscous destruction of enstrophy (16) Note that the term that we have called turbulent diffusion of the small-scale enstrophy should really be written as ω li τ l (ω k,u i ) = ω li τ l (ω i,u k ) + ω li (τ l (ω k,u i ) τ l (ω i,u k )) small-scale advection contribution turbulent diffusion proper Where the vortex-force term τ l (u i,ω j ) τ l (u j,ω i ) appears above, we could have used also the subscale force term ɛ ijk flk s. Other forms of the relation may also be written. For example, again using the relations j τ l (ω j,u i )=τ l (ω j,u i,j )=τ l (ω j,s ij ), the small-scale enstrophy balance may be written as t ζ l + k [ζ l ū l,k τ l(ω i,ω i,u k ) ν k ζ l ] = τ l (S ij,ω i,ω j )+τ l (ω i,ω j )S li,j + τ l (ω i,s ij ) ω lj (17) ω li,k τ l (ω i,u k ) ντ l (ω i,k,ω i,k ) (18) This is the version presented in T & L, eq.(3.3.38). Note that the production of enstrophy in the small-scales by flux from the large-scales ω li,j (τ l (ω i,u j ) τ l (ω j,u i )) [or the similar term that appears in the T & L versions] is exactly equal and opposite to the enstrophy flux out of the large-scales in (13). This is very similar to what occurred in the energy balances. However, we shall see that the enstrophy dynamics is NOT dominated by nonlinear cascade! 17