betaFeedback

WorksPhysics and Astronomy

On the transport equation for probability density functions of turbulent vorticity fields

Abstract

The vorticity random field of turbulent flow is singled out as the main dynamical variable for the description of turbulence, and the evolution equation of the probability density function (PDF) of the vorticity field has been obtained. This PDF evolution equation is a mixed type partial differential equation (PDE) of second order which depends only on the conditional mean (first order) of the underlying turbulent flow, which is in contrast with Reynolds' mean flow equation which relies on a quadratic statistics. Therefore the new PDF PDE may provide new closure scheme based on the conditional linear statistics, and some of them will be described in the present paper too.

Introduction

In statistical fluid mechanics (cf. [13]), the velocity U(x,t)U(x,t) of a turbulent flow is promoted to a random field, cf. [8], indexed by space variable x∈R3x \in\mathbb{R}^{3} and time parameter tt. From this point of view, the turbulence problem, if there is one, seeks for a description of the distribution of the velocity field. This distribution is rather complicated and consists of all joint distributions of the velocity across over finitely many locations and times, and therefore it is challenging to describe the distributions of turbulent flows in general. As early as in 1950’s, E. Hopf [6] (cf. [13] too) made an ambitious attempt and derived differential equations for the distribution of a turbulent flow. Hopf’s differential equations are however infinite dimensional and involve functional derivatives, which are therefore too difficult to extract useful information about turbulence. The dominant approach in the statistical theory of turbulence, initiated in G. I. Taylor’s seminal work [18, 19], has been based on the analysis of moment structure functions and relied on the spectral method (cf. [1, 13] for example). In the past decades, attention has been paid to the one-dimensional marginal distribution of the velocity field U(x,t)U(x,t), whose distribution is finite-dimensional. It is reasonable to assume that the distribution of U(x,t)U(x,t), with (x,t)(x,t) fixed, has a probability density function (PDF) with respect to the Lebesgue measure on R3\mathbb{R}^3. The PDF methods based on formal PDF transport equations have been developed through a series of work by Pope and other researchers (cf. [14, 15]), which become powerful tools for modeling turbulent flows. The PDF, by definition, has to satisfy “the adjoint equation” of the fluid dynamic equations. If the turbulent flow is an incompressible viscous fluid flow, then the PDF of the velocity must satisfy the adjoint equation of the Navier-Stokes equation. The transport equation for PDF known in the literature is still a formal adjoint equation of the Navier-Stokes equation, and therefore only few features from the transport equation may be used in modeling turbulent flows.

For incompressible fluid flows, the Navier-Stokes equation is equivalent to the vorticity equation (see (2) below), and therefore it is natural to consider the vorticity W=∇∧UW = \nabla\wedge U as the main fluid dynamic variable for the study of turbulence [12]. There are good reasons why we should concentrate on the vortex motion in turbulence. The velocity of a turbulent flow is unlikely to be independently or possess conditional independence with respect to the spatial variable, while vortex motions of many turbulent flows observed in nature (such as vortex lines, vortex rings) acquire certain conditional independence, in the sense that by focusing on the motion near a fixed region, the future vortex motions evolve more or less independent of what happens in other positions. In fact there are good evidences which demonstrate that some sort of superposition property of vorticity may be maintained, although not exactly due to highly non-linear and non-local nature of turbulence. These observations are valuable in modeling turbulent flows via the vorticity, which are already applied in vortex methods (cf. [3, 11]). In this sense PDF methods based on the vorticity are valuable.

The main contribution of the present work is the partial differential equation for the PDF of the vorticity (called PDF PDE, or PDF equation for short), which will be derived in the main body of the paper. The PDF PDE is highly non-linear, however the most striking aspect is that the PDF PDE for the vorticity depends only on one single first-order statistical characteristic of the turbulent flow. More precisely, we identify the main statistical characteristic needed for the PDF PDE with the conditional mean function of the increment of the vorticity given the current vorticity:

μi(x,y,w,t)=E[Wi(y,t)−Wi(x,t)∣W(x,t)=w](1)\mu^i(x,y,w,t) = \mathbb{E}\left[W^i(y,t)-W^i(x,t)\mid W(x,t)=w\right] \tag*{(1)}

for i=1,2,3i = 1,2,3, where x,y,w∈R3x,y,w \in\mathbb{R}^3 and t≥0t \ge0. The PDF of the vorticity WW is a solution to the PDF PDE, which is a second-order partial differential equation where coefficients appearing in the PDF PDE depend on μ\mu only. Besides its theoretical interest, this PDF PDE paves the way towards practical modeling the statistics of turbulent flows based on the vorticity PDF.

The paper is organized as the following. In Section 2, several notions and notations together with several standard assumptions about fluid dynamic random fields will be introduced, and the main result, namely the PDF PDE for vorticity will be derived. In Sections 3 and 4, the theoretical foundation, based on our PDF PDE for vorticity, will be laid for the purpose of modeling various turbulent flows. In Section 5, we propose the most direct way of modeling PDF of the vorticity by specifying the conditional mean function μi\mu^i. The PDF PDE considered purely as a PDE theory is over-determined, due to its non-linearity in the sense that the coefficients appearing in the PDF PDE are not independent of its solutions. Therefore care is needed to ensure that the additional constraint is satisfied. In the last section, we propose the heat flow method to model statistical quantities needed for closing the PDF PDE and obtain concrete PDF examples.

Conventions on notations. The following conventions are used throughout the paper. Firstly Einstein’s convention on summation on repeated indices through their ranges is assumed, unless otherwise specified. If AA is a vector or a vector field (in the space of dimension three) dependent on some parameters, then its components are labeled with upper-script indices, i.e. A=(Ai)=(A1,A2,A3)A=(A^i)=(A^1,A^2,A^3). The same convention applies to coordinates too. Partial derivatives of functions may be labeled with variables in sub-scripts. For example, if A(w;x,t)A(w;x,t) is a vector-valued function depending on w,x∈R3w,x\in\mathbb{R}^3 and tt, then ∇wA\nabla_w A means the total derivatives (∂∂wiAj)\left(\frac{\partial}{\partial w^i}A^j\right), ΔxA\Delta_x A means the vector (∂2∂xk∂xkAi)\left(\frac{\partial^2}{\partial x^k\partial x^k}A^i\right) of the Laplacians of AiA^i. However, as a general rule, derivatives without sub-scripts mean the derivatives with respect to the variable x=(xi)x=(x^i), unless otherwise specified for avoiding possible confusion.

The velocity vector field will be denoted by UU. W=∇∧UW=\nabla\wedge U is its vorticity so that Wi=εijk∂∂xjUkW^i=\varepsilon^{ijk}\frac{\partial}{\partial x^j}U^k, where εijk\varepsilon^{ijk} are the Levi-Civita symbols.

PDF equation for the vorticity

In this paper we regard the vorticity of an incompressible turbulent flow as the main dynamic variable. The goal of this section is to derive the evolution equation for the PDF of the vorticity of an incompressible turbulent flow.

Prelims and assumptions

Let U(x,t)U(x,t) be the velocity of an incompressible turbulent flow in R3\mathbb{R}^3 with viscosity ν\nu. Suppose there is no external force supplied to the turbulence. Being a random field though, U(x,t)U(x,t) satisfies the Navier-Stokes equations

∂Ui∂t+Uj∂Ui∂xj=νΔUi−∂P∂xi\frac{\partial U^i}{\partial t}+U^j\frac{\partial U^i}{\partial x^j}=\nu\Delta U^i-\frac{\partial P}{\partial x^i}

and

∂Uj∂xj=0,\frac{\partial U^j}{\partial x^j}=0,

where i=1,2,3i=1,2,3, and PP is the pressure, subject to initial conditions which are random. The motion equations of the vorticity W=∇∧UW=\nabla\wedge U are the vorticity equations

∂Wi∂t+Uj∂Wi∂xj=νΔWi+Wj∂Ui∂xj(2)\frac{\partial W^i}{\partial t}+U^j\frac{\partial W^i}{\partial x^j}=\nu\Delta W^i+W^j\frac{\partial U^i}{\partial x^j} \tag*{(2)}

for i=1,2,3i=1,2,3.

We make the following technical assumptions.

First, we assume that both U(x,t)U(x,t) and W(x,t)W(x,t) have derivatives in xx and tt of any order, and theses derivatives decay to zero at infinity sufficiently fast, so that possible boundary terms arising in applications of the Stokes’ formula have no contributions in computations below. Therefore, since UU is divergence-free, ΔU=−∇∧W\Delta U=-\nabla\wedge W, according to Green’s formula

Ui(x,t)=∫R314π∣x−y∣εijk∂∂yjWk(y,t) dyU^i(x,t)=\int_{\mathbb{R}^3}\frac{1}{4\pi\lvert x-y\rvert}\varepsilon^{ijk}\frac{\partial}{\partial y^j}W^k(y,t)\,\mathrm{d}y

which yields the Biot-Savart law:

Ui(x,t)=−∫R3εijkxj−yj4π∣x−y∣3Wk(y,t) dy.(3)U^i(x,t)=-\int_{\mathbb{R}^3}\varepsilon^{ijk}\frac{x^j-y^j}{4\pi\lvert x-y\rvert^3}W^k(y,t)\,\mathrm{d}y. \tag*{(3)}

for i=1,2,3i=1,2,3.

Our second technical assumption is to impose certain regularity on the distribution of the vorticity. At each point xx and instance t≥0t \ge0, W(x,t)W(x,t) is a random variable defined on a probability space (Ω,F,P)(\Omega,\mathcal{F},\mathbb{P}), taking values in R3\mathbb{R}^{3}, and therefore its law (or called distribution) is a probability measure on (R3,B(R3))(\mathbb{R}^{3},\mathcal{B}(\mathbb{R}^{3})). We assume that W(x,t)W(x,t) has a positive and smooth probability density function (PDF), denoted by f(w;x,t)f(w;x,t), in the sense that

P(W(x,t)∈A)=∫Af(w;x,t) dw\mathbb{P}(W(x,t) \in A) = \int_A f(w;x,t)\,\mathrm{d}w

for every Borel measurable subset AA. The joint distribution of W(x,t)W(x,t) and W(y,t)W(y,t) for any pair x≠yx \ne y has a positive and smooth PDF, denoted by f2(w1,w2;x,y,t)f_2(w_1,w_2;x,y,t).

To state our third assumption, we first notice that

f2∣1(w;w1,x,y,t)=f2(w1,w;x,y,t)f(w1;x,t)f_{2|1}(w;w_1,x,y,t) = \frac{f_2(w_1,w;x,y,t)}{f(w_1;x,t)}

is the PDF of the conditional law of W(y,t)W(y,t) given W(x,t)=w1W(x,t)=w_1. The conditional mean function μi(x,y,w,t)\mu^i(x,y,w,t) is defined by

μi(x,y,w,t)=E[Wi(y,t)−Wi(x,t)∣W(x,t)=w]=∫R3(w1i−wi)f2∣1(w1;w,x,y,t) dw1(4)\begin{aligned} \mu^i(x,y,w,t) &= \mathbb{E}\left[W^i(y,t)-W^i(x,t)\mid W(x,t)=w\right] \\ &= \int_{\mathbb{R}^{3}}(w_1^i-w^i)f_{2|1}(w_1;w,x,y,t)\,\mathrm{d}w_1 \tag*{(4)} \end{aligned}

for i=1,2,3i=1,2,3, which will play a dominant role in the sequel.

The third technical assumption is about the regularity of the conditional average function μ=(μ1,μ2,μ3)\mu=(\mu^1,\mu^2,\mu^3). It is assumed that the derivatives of μ\mu of any order exist and decay to zero sufficiently fast at infinity. Moreover it is assumed that μ\mu has an asymptotic expansion

μi(x,y,w,t)=aki(x,w,t)(yk−xk)+bjki(x,w,t)(yk−xk)(yj−xj)+o(∣y−x∣2)(5)\mu^i(x,y,w,t)=a_k^i(x,w,t)(y^k-x^k)+b_{jk}^i(x,w,t)(y^k-x^k)(y^j-x^j)+o\left(|y-x|^2\right) \tag*{(5)}

as lverty−x∣→0lvert y-x\rvert\to0, where akia_k^i's and bjkib_{jk}^i's are assumed to be continuous with respect to all of their arguments. We denote bi=bkkib^i=b_{kk}^i for i=1,2,3i=1,2,3. It is clear that

aki(x,w,t)=E[∂Wi∂xk(x,t)∣W(x,t)=w]=∂∂ykμi(x,y,w,t)∣y=x(6)a_k^i(x,w,t)=\mathbb{E}\left[\left.\frac{\partial W^i}{\partial x^k}(x,t)\right|W(x,t)=w\right]=\left.\frac{\partial}{\partial y^k}\mu^i(x,y,w,t)\right|_{y=x} \tag*{(6)}

which represents the local rate of change in the vortex motion over the turbulent region, and

bi(x,w,t)=12E[ΔxWi(x,t)∣W(x,t)=w]=12Δyμi(x,y,w,t)∣y=x.(7)b^i(x,w,t)=\frac{1}{2}\mathbb{E}\left[\left.\Delta_x W^i(x,t)\right|W(x,t)=w\right]=\left.\frac{1}{2}\Delta_y\mu^i(x,y,w,t)\right|_{y=x}. \tag*{(7)}

In the remainder of the paper, we will work with a turbulent flow for which the three assumptions listed above are satisfied.

PDF equation and its derivation

In this section, we derive the main result of the paper, that is, a partial differential equation which the PDF of the vorticity must satisfy.

Theorem 1. Under the assumptions and notations established in Sec. 2.1, suppose the PDF f(w;x,t)f(w;x,t) of the vorticity is smooth in (w,x,t)(w,x,t) and has finite moments, that is ∫R3∣w∣nf(w;x,t) dw<∞\int_{\mathbb{R}^{3}}|w|^n f(w;x,t)\,\mathrm{d}w<\infty for every n=1,2,…n=1,2,\ldots. Then ff satisfies the following partial differential equation:

(∂∂t+∂Bi∂xi+Bi∂∂xi−νΔx)f=ν∂∂wi(∂∂xk(faki)−2bif)+∂∂wi(fDi)(8)\left(\frac{\partial}{\partial t}+\frac{\partial B^i}{\partial x^i}+B^i\frac{\partial}{\partial x^i}-\nu\Delta_x\right)f = \nu\frac{\partial}{\partial w^i}\left(\frac{\partial}{\partial x^k}(fa_k^i)-2b^if\right) +\frac{\partial}{\partial w^i}(fD^i) \tag*{(8)}

where Δx\Delta_x is the Laplacian with respect to the space variable xx, aka_k and bb are given as in (6) and (7),

Bi(x,w,t)=∫R314π∣y−x∣εijk∂∂yjμk(x,y,w,t) dy.(9)B^i(x,w,t)=\int_{\mathbb{R}^3}\frac{1}{4\pi\lvert y-x\rvert}\varepsilon^{ijk}\frac{\partial}{\partial y^j}\mu^k(x,y,w,t)\,dy. \tag*{(9)}

and

Di(x,w,t)=wl∫R3yl−xl4π∣y−x∣3εijk∂∂yjμk(x,y,w,t) dy.(10)D^i(x,w,t)=w^l\int_{\mathbb{R}^3}\frac{y^l-x^l}{4\pi\lvert y-x\rvert^3}\varepsilon^{ijk}\frac{\partial}{\partial y^j}\mu^k(x,y,w,t)\,dy. \tag*{(10)}

Proof. Let FF be a smooth function on R3\mathbb{R}^3 with a compact support. By the definition of PDF one has

∫R3F(w)∂∂tf(w;x,t) dw=E[∂∂tF(W(x,t))].\int_{\mathbb{R}^3}F(w)\frac{\partial}{\partial t}f(w;x,t)\,dw=\mathbb{E}\left[\frac{\partial}{\partial t}F(W(x,t))\right].

We are going to calculate the right-hand side expectation in terms of the PDF f(w;x,t)f(w;x,t) and other statistical characteristics of the turbulent flow. This will be done by using the following equation

∂∂tF(W(x,t))=νΔxF(W)−∂(UiF(W))∂xi+Fj(W)Wi∂Uj∂xi−ν∂Fj(W)∂xk∂Wj∂xk,\frac{\partial}{\partial t}F(W(x,t))=\nu\Delta_xF(W)-\frac{\partial\left(U^iF(W)\right)}{\partial x^i}+F_j(W)W^i\frac{\partial U^j}{\partial x^i}-\nu\frac{\partial F_j(W)}{\partial x^k}\frac{\partial W^j}{\partial x^k},

where, for simplicity, FjF_j denote the partial derivatives ∂F∂xj\frac{\partial F}{\partial x^j} of FF, j=1,2,3j=1,2,3. This equation follows directly from the vorticity equations. From the previous equation we obtain that

E[∂∂tF(W(x,t))]=νE[ΔxF(W)]−E[∂(UiF(W))∂xi]+E[Fj(W)Wi∂Uj∂xi]−E[ν∂Fj(W)∂xk∂Wj∂xk]\mathbb{E}\left[\frac{\partial}{\partial t}F(W(x,t))\right]=\nu\mathbb{E}\left[\Delta_xF(W)\right]-\mathbb{E}\left[\frac{\partial\left(U^iF(W)\right)}{\partial x^i}\right] +\mathbb{E}\left[F_j(W)W^i\frac{\partial U^j}{\partial x^i}\right]-\mathbb{E}\left[\nu\frac{\partial F_j(W)}{\partial x^k}\frac{\partial W^j}{\partial x^k}\right]
:=I1+I2+I3+I4.(11):=I_1+I_2+I_3+I_4. \tag*{(11)}

Let us calculate IiI_i’s on the right-hand side of (11). First, it is easy to see that by definition

I1=νΔxE[F(W)]=∫R3F(w)νΔxf(w;x,t) dw.(12)I_1=\nu\Delta_x\mathbb{E}[F(W)]=\int_{\mathbb{R}^3}F(w)\nu\Delta_xf(w;x,t)\,dw. \tag*{(12)}

For computing I2I_2, we shall use (3) and obtain that

I2=−E[∂(UiF(W))∂xi]=−∂∂xiE[F(W)Ui]=−∂∂xiE[F(W)εijk∫R314π∣y−x∣∂∂yjWk(y,t) dy].\begin{aligned} I_2=-\mathbb{E}\left[\frac{\partial\left(U^iF(W)\right)}{\partial x^i}\right]=-\frac{\partial}{\partial x^i}\mathbb{E}\left[F(W)U^i\right] \\ &=-\frac{\partial}{\partial x^i}\mathbb{E}\left[F(W)\varepsilon^{ijk}\int_{\mathbb{R}^3}\frac{1}{4\pi\lvert y-x\rvert}\frac{\partial}{\partial y^j}W^k(y,t)\,dy\right]. \end{aligned}

To work out the expectation on the right-hand side, we may rewrite the partial derivative as a limit:

∂∂yjWk(y,t)=lim⁡h→01h(Wk(y+he(j),t)−Wk(y,t)),\frac{\partial}{\partial y^j}W^k(y,t)=\lim_{h\to0}\frac{1}{h}\left(W^k(y+he^{(j)},t)-W^k(y,t)\right),

where e(j)e^{(j)} represents the unit vector with jj-th component equal to 1, and the rest two components 0, so that we can rewrite I2I_2 as the following limit

I2=−∂∂xilim⁡h→01hE[F(W)εijk∫R314π∣y−x∣(Wk(y+he(j),t)−Wk(y,t)) dy].(13)I_2=-\frac{\partial}{\partial x^i}\lim_{h\to0}\frac{1}{h}\mathbb{E}\left[F(W)\varepsilon^{ijk}\int_{\mathbb{R}^3}\frac{1}{4\pi\lvert y-x\rvert}\left(W^k(y+he^{(j)},t)-W^k(y,t)\right)\,dy\right]. \tag*{(13)}

The expectation in this expression may be written in terms of two-point joint distributions of WW as the following

∫R3F(w)[∫εijk4π∣y−x∣(∫R3w1k(f2(w,w1;x,y+he(j),t)−f2(w,w1;x,y,t)) dw1) dy] dw.\int_{\mathbb{R}^{3}} F(w)\left[\int\frac{\varepsilon^{ijk}}{4\pi|y-x|}\left(\int_{\mathbb{R}^{3}}w_{1}^{k}\left(f_{2}(w,w_{1};x,y+he^{(j)},t)-f_{2}(w,w_{1};x,y,t)\right)\,\mathrm{d}w_{1}\right)\,\mathrm{d}y\right]\,\mathrm{d}w.

The inner integral against the variable w1w_{1} equals

f(w;x,t)(μ(x,y+he(j),w,t)−μ(x,y,w,t)).f(w;x,t)\left(\mu(x,y+he^{(j)},w,t)-\mu(x,y,w,t)\right).

After substituting this in Eq. (13) and sending h→0h\to0, we obtain that

I2=∫R3F(w)[−∂∂xi(f(w;x,t)∫R3εijk4π∣y−x∣∂∂yjμk(x,y,w,t) dy)] dw.(14)I_{2}=\int_{\mathbb{R}^{3}}F(w)\left[-\frac{\partial}{\partial x^{i}}\left(f(w;x,t)\int_{\mathbb{R}^{3}}\frac{\varepsilon^{ijk}}{4\pi|y-x|}\frac{\partial}{\partial y^{j}}\mu^{k}(x,y,w,t)\,\mathrm{d}y\right)\right]\,\mathrm{d}w. \tag*{(14)}

Next we deal with I3I_{3}. Again, using (3), we may write

∂∂xjUi(x,t)=∫R3yj−xj4π∣y−x∣3εilk∂∂ylWk(y,t) dy.\frac{\partial}{\partial x^{j}}U^{i}(x,t)=\int_{\mathbb{R}^{3}}\frac{y^{j}-x^{j}}{4\pi|y-x|^{3}}\varepsilon^{ilk}\frac{\partial}{\partial y^{l}}W^{k}(y,t)\,\mathrm{d}y.

so that

I3=E[Fi(W)Wj∂Ui∂xj]=E[Fi(W)Wj∫R3yj−xj4π∣y−x∣3εilk∂∂ylWk(y,t) dy],\begin{aligned} I_{3}=\mathbb{E}\left[F_{i}(W)W^{j}\frac{\partial U^{i}}{\partial x^{j}}\right] \\ &=\mathbb{E}\left[F_{i}(W)W^{j}\int_{\mathbb{R}^{3}}\frac{y^{j}-x^{j}}{4\pi|y-x|^{3}}\varepsilon^{ilk}\frac{\partial}{\partial y^{l}}W^{k}(y,t)\,\mathrm{d}y\right], \end{aligned}

which can be evaluated by using the two-point joint PDF. Indeed, we may repeat the same idea as in the computation of I2I_{2}, to obtain that

I3=∫R3F(w)∂∂wi[−f(w;x,t)εilkwj∫R3yj−xj4π∣y−x∣3∂∂ylμk(x,y,w,t) dy] dw.(15)I_{3}=\int_{\mathbb{R}^{3}}F(w)\frac{\partial}{\partial w_{i}}\left[-f(w;x,t)\varepsilon^{ilk}w_{j}\int_{\mathbb{R}^{3}}\frac{y^{j}-x^{j}}{4\pi|y-x|^{3}}\frac{\partial}{\partial y^{l}}\mu^{k}(x,y,w,t)\,\mathrm{d}y\right]\,\mathrm{d}w. \tag*{(15)}

Similarly, for the last term I4I_{4}, we write

I4=−ν∑k=13E[∂Fi(W)∂xk∂Wi∂xk]=−ν∑k=13lim⁡h→01h2E[(Fi(W(x+he(k),t))−Fi(W(x,t)))(Wi(x+he(k),t)−Wi(x,t))]=−ν∑k=13lim⁡h→01h2∬R3×R3(Fi(w2)−Fi(w1))(w2i−w1i)f2(w1,w2,x,x+he(k),t) dw1dw2=ν∫R3Fi(w)lim⁡h→01h2∑k=13{f(w;x+he(k),t)∫R3(w1i−wi)f2∣1(w1;w,x+he(k),x,t) dw1\begin{aligned} I_{4}=-\nu\sum_{k=1}^{3}\mathbb{E}\left[\frac{\partial F_{i}(W)}{\partial x^{k}}\frac{\partial W^{i}}{\partial x^{k}}\right] \\ &=-\nu\sum_{k=1}^{3}\lim_{h\to0}\frac{1}{h^{2}}\mathbb{E}\left[\left(F_{i}(W(x+he^{(k)},t))-F_{i}(W(x,t))\right)\left(W^{i}(x+he^{(k)},t)-W^{i}(x,t)\right)\right] \\ &=-\nu\sum_{k=1}^{3}\lim_{h\to0}\frac{1}{h^{2}}\iint_{\mathbb{R}^{3}\times\mathbb{R}^{3}}\left(F_{i}(w_{2})-F_{i}(w_{1})\right)\left(w_{2}^{i}-w_{1}^{i}\right)f_{2}(w_{1},w_{2},x,x+he^{(k)},t)\,\mathrm{d}w_{1}\mathrm{d}w_{2} \\ &=\nu\int_{\mathbb{R}^{3}}F_{i}(w)\lim_{h\to0}\frac{1}{h^{2}}\sum_{k=1}^{3}\left\{f(w;x+he^{(k)},t)\int_{\mathbb{R}^{3}}\left(w_{1}^{i}-w^{i}\right)f_{2|1}(w_{1};w,x+he^{(k)},x,t)\,\mathrm{d}w_{1} \right. \end{aligned}
+f(w;x,t)∫R3(w1i−wi)f2∣1(w1;w,x,x+he(k),t) dw1} dw=ν∫R3Fi(w)lim⁡h→01h2I4h dw,\begin{aligned} \left. +f(w;x,t)\int_{\mathbb{R}^{3}}\left(w_{1}^{i}-w^{i}\right)f_{2|1}(w_{1};w,x,x+he^{(k)},t)\,\mathrm{d}w_{1}\right\}\,\mathrm{d}w \\ &=\nu\int_{\mathbb{R}^{3}}F_{i}(w)\lim_{h\to0}\frac{1}{h^{2}}I_{4}^{h}\,\mathrm{d}w, \end{aligned}

where for simplicity we have introduced the notation

I4h:=∑k=13{(f(w;x+he(k),t)−f(w;x,t))μi(x+he(k),x,w,t)+f(w;x,t)(μi(x,x+he(k),w,t)+μi(x+he(k),x,w,t))}.I_4^h := \sum_{k=1}^{3} \left\{ \left(f(w;x+he^{(k)},t)-f(w;x,t)\right)\mu^i(x+he^{(k)},x,w,t) + f(w;x,t)\left(\mu^i(x,x+he^{(k)},w,t)+\mu^i(x+he^{(k)},x,w,t)\right) \right\}.

I4hI_4^h can be evaluated by using (5) when hh is sufficiently small, by which we mean that

μi(x,x+he(k),w,t)=aki(x,w,t)h+bkki(x,w,t)h2+o(h2).\mu^i(x,x+he^{(k)},w,t)=a_k^i(x,w,t)h+b_{kk}^i(x,w,t)h^2+o(h^2).

and

μi(x+he(k),x,w,t)=−aki(x+he(k),w,t)h+bkki(x+he(k),w,t)h2+o(h2).\mu^i(x+he^{(k)},x,w,t)=-a_k^i(x+he^{(k)},w,t)h+b_{kk}^i(x+he^{(k)},w,t)h^2+o(h^2).

Consequently, we have

μi(x,x+he(k),w,t)+μi(x+he(k),x,w,t)=−(aki(x+he(k),w,t)−aki(x,w,t))h+(bkki(x,w,t)+bkki(x+he(k),w,t))h2+o(h2),\mu^i(x,x+he^{(k)},w,t)+\mu^i(x+he^{(k)},x,w,t)=-\left(a_k^i(x+he^{(k)},w,t)-a_k^i(x,w,t)\right)h+\left(b_{kk}^i(x,w,t)+b_{kk}^i(x+he^{(k)},w,t)\right)h^2+o(h^2),

which yields that

lim⁡h→01h2I4h=−∂∂xk(faki)+2f∑kbkki.\lim_{h\to0}\frac{1}{h^2}I_4^h=-\frac{\partial}{\partial x^k}(fa_k^i)+2f\sum_k b_{kk}^i.

and we may conclude that

I4=−ν∫R3Fi(w)[∂∂xk(faki)−2f∑kbkki] dw=∫R3F(w)∂∂wi[ν∂∂xk(faki)−2νf∑kbkki] dw.(16)\begin{aligned} I_4=-\nu\int_{\mathbb{R}^3}F_i(w)\left[\frac{\partial}{\partial x^k}(fa_k^i)-2f\sum_k b_{kk}^i\right]\,dw \\ &=\int_{\mathbb{R}^3}F(w)\frac{\partial}{\partial w^i}\left[\nu\frac{\partial}{\partial x^k}(fa_k^i)-2\nu f\sum_k b_{kk}^i\right]\,dw. \tag*{(16)} \end{aligned}

The PDF equation (8) now follows by substituting (12, 14, 15, 16) into (11).

Remark 2. The partial differential equation (8), although it appears linear, is in fact highly non-linear, and more importantly, the coefficients a,b,Ba,b,B and DD that define the partial differential equation cannot be in general determined by the PDF ff alone. We have seen that these functions are functionals of the conditional average function μ\mu, hence the PDF ff cannot be obtained solely through the PDF equation. However, the PDF PDE provides an appealing and new method for modeling the PDF ff based on the modeling of μ\mu alone, we will provide in this work several results on modeling the PDF ff based on the PDF equation we just derived.

Remark 3. If all coefficients a,b,Ba,b,B and DD are considered as given, then we can pose the initial value problem for solving the PDF PDE (8). The good news is that then the PDF PDE is a scalar linear partial differential equation of second order, although in general it is a mixed type of partial differential equations of second order on 6 dimensional space. To the best knowledge of the authors of the present paper, this kind of partial differential equations has not been studied systemically in the existing literature.

The distribution of turbulence

As an application of the PDF PDE obtained in the previous sub-section, we address a longstanding question in turbulence about the distribution of turbulent flows. It has been conjectured and verified by measurements over many years (cf. [1] and [4]) that the distribution of a genuine turbulent flow cannot be Gaussian, while a mathematical proof for this statement, to the best knowledge of the present authors, is not available yet. By using the PDF PDE we are able to prove the following

Theorem 4. Consider an incompressible viscous (with viscosity ν>0\nu> 0) turbulent flow with vorticity W(x,t)\mathbf{W}(x,t). Suppose

  1. the mean vorticity is constant at any instance,

  1. {W(x,t)}\{\mathbf{W}(x,t)\} is weakly isotropic in the sense that

E[∇W(x,t)∣W(x,t)=w]=0\mathbb{E}[\nabla\mathbf{W}(x,t)\mid\mathbf{W}(x,t)=w]=0

for all x,wx,w and t≥0t \ge0, and

  1. the distribution of {W(x,t)}\{\mathbf{W}(x,t)\} is Gaussian.

Then the distribution of W(x,t)\mathbf{W}(x,t) is independent of xx.

Proof. Since {W(x,t):x∈Rd and t≥0}\{\mathbf{W}(x,t):x\in\mathbb{R}^{d}\text{ and }t\ge0\} is a Gaussian random field, for x≠yx\ne y, the joint law of W(x,t)\mathbf{W}(x,t) and W(y,t)\mathbf{W}(y,t) is a normal distribution on R6\mathbb{R}^{6} whose co-variance matrix Σ(x,y)\Sigma(x,y) may be decomposed into blocks

Θ(x,y)=(ΣxΣx,yΣy,xΣy),\Theta(x,y)=\begin{pmatrix}\Sigma_x & \Sigma_{x,y}\\ \Sigma_{y,x} & \Sigma_y\end{pmatrix},

where the dependence on tt is suppressed for simplicity. The PDF of W(x,t)\mathbf{W}(x,t) is given by

f(w;x,t)=1(2π)3/2det⁡Σxexp⁡(−12(w−m)TΣx−1(w−m)).f(w;x,t)=\frac{1}{(2\pi)^{3/2}\sqrt{\det\Sigma_x}}\exp\left(-\frac{1}{2}(w-m)^{T}\Sigma_x^{-1}(w-m)\right).

where mm denotes the mean vorticity, so that

ln⁡f(w;x,t)=−32ln⁡(2π)−12ln⁡det⁡Σx−12(w−m)TΣx−1(w−m).(17)\ln f(w;x,t)=-\frac{3}{2}\ln(2\pi)-\frac{1}{2}\ln\det\Sigma_x-\frac{1}{2}(w-m)^{T}\Sigma_x^{-1}(w-m). \tag*{(17)}

Since under the assumption aki=0a_k^i=0, ψ=ln⁡f\psi=\ln f according to PDF PDE (8) must satisfy the following equation:

∂ψ∂t=νΔxψ−Bi∂ψ∂xi+ν∣∇xψ∣2+(Di−2νbi)∂ψ∂wi+∂(Di−2νbi)∂wi−∇x⋅B\frac{\partial\psi}{\partial t}=\nu\Delta_x\psi-B^i\frac{\partial\psi}{\partial x^i}+\nu\lvert\nabla_x\psi\rvert^2+\left(D^i-2\nu b^i\right)\frac{\partial\psi}{\partial w^i}+\frac{\partial\left(D^i-2\nu b^i\right)}{\partial w^i}-\nabla_x\cdot B

(which is the PDF PDE for ψ=ln⁡f\psi=\ln f in the case where akia_k^i vanish). Using (17) and by a lengthy but completely elementary computation, we deduce that

−2∂ψ∂t=νΔxln⁡det⁡Σx−12ν∣∇xln⁡det⁡Σx∣2+4ν∂bi∂wi−Bi∂ln⁡det⁡Σx∂xi−2∂Di∂wi+2∇x⋅B+νuTΔxΣx−1u−νuT(∑i∂ln⁡det⁡Σx∂xi∂Σx−1∂xi)u−4νbi(Σx−1)iβuβ+2Di(Σx−1)iβuβ−BiuT∂Σx−1∂xiu−12ν∣uT∇xΣx−1u∣2\begin{aligned} -2\frac{\partial\psi}{\partial t}=\nu\Delta_x\ln\det\Sigma_x-\frac{1}{2}\nu\lvert\nabla_x\ln\det\Sigma_x\rvert^2+4\nu\frac{\partial b^i}{\partial w^i} -B^i\frac{\partial\ln\det\Sigma_x}{\partial x^i}-2\frac{\partial D^i}{\partial w^i}+2\nabla_x\cdot B+\nu u^T\Delta_x\Sigma_x^{-1}u -\nu u^T\left(\sum_i\frac{\partial\ln\det\Sigma_x}{\partial x^i}\frac{\partial\Sigma_x^{-1}}{\partial x^i}\right)u-4\nu b^i(\Sigma_x^{-1})_{i\beta}u^\beta\\ &+2D^{i}(\Sigma_x^{-1})_{i\beta}u^\beta-B^i u^T\frac{\partial\Sigma_x^{-1}}{\partial x^i}u-\frac{1}{2}\nu\left|u^T\nabla_x\Sigma_x^{-1}u\right|^2 \end{aligned}

where u=w−mu=w-m. The last equation looks complicated, but the important point in our argument is the observation that, the left-hand side −2∂ψ∂t-2\frac{\partial\psi}{\partial t} is a quadratic function of uu, while the right-hand side is a polynomial in uu of degree 4, and only the last term on the right-hand side has order 4 in uu. Therefore

12ν∣uT∇xΣx−1u∣2=0\frac{1}{2}\nu\left|u^T\nabla_x\Sigma_x^{-1}u\right|^2=0

for every uu, which yields that ∇xΣx−1=0\nabla_x\Sigma_x^{-1}=0, so that Σx\Sigma_x is independent of xx. This completes the proof.

Inviscid fluid flows

There is a significant simplification in PDF PDE for an inviscid “turbulent” flow, although such turbulence may not exist in nature. For inviscid fluid flows, the velocity U(x,t)U(x,t) satisfies the Euler equations

∂∂tUi+Uj∂Ui∂xj=−∇P,∂Ui∂xi=0.\frac{\partial}{\partial t}U^i+U^j\frac{\partial U^i}{\partial x^j}=-\nabla P,\qquad\frac{\partial U^i}{\partial x^i}=0.

The PDF f(w;x,t)f(w;x,t) of W=∇∧UW=\nabla\wedge U satisfies the (non-linear) transport differential equation

∂f∂t+∂∂xi(fBi)=∂∂wi(fDi),(18)\frac{\partial f}{\partial t}+\frac{\partial}{\partial x^i}(fB^i)=\frac{\partial}{\partial w^i}(fD^i), \tag*{(18)}

where BB and DD are given as in (9) and (10) respectively.

The following theorem provides a mathematical tool for modeling PDF of an inviscid turbulent flow.

Theorem 5. Consider PDE (18) where the data BB and DD are assumed as given. Assume that BB and DD are Lipschitz continuous in (x,w)(x,w) uniformly in t>0t>0. Suppose f(w;x,t)f(w;x,t) is a solution to (18) with continuous initial data f(w;x,0)f(w;x,0), and suppose fDfD decays to zero sufficiently fast as ∣w∣→∞|w|\to\infty. Then we have the followings:

  1. f(w;x,0)≥0f(w;x,0)\ge0 for all x∈R3x\in\mathbb{R}^3, then f(w;x,t)≥0f(w;x,t)\ge0 for all x∈R3x\in\mathbb{R}^3 and t>0t>0.

  1. Suppose ∫R3f(w;x,0) dw=1\int_{\mathbb{R}^3}f(w;x,0)\,dw=1 for all x∈R3x\in\mathbb{R}^3, then ∫R3f(w;x,t) dw=1\int_{\mathbb{R}^3}f(w;x,t)\,dw=1 for all x∈R3x\in\mathbb{R}^3 and t>0t>0, if and only if

∂∂xi∫R3Bi(x,w,t)f(w;x,t) dw=0(19)\frac{\partial}{\partial x^i}\int_{\mathbb{R}^3}B^i(x,w,t)f(w;x,t)\,dw=0 \tag*{(19)}

for all x∈R3x\in\mathbb{R}^3 and $t>0.

Proof. Let us define the integral curves XX and YY by the following ordinary differential equation system:

{ddsXi(t,s)=−Bi(X(t,s),Y(t,s),t−s)),X(t,0)=x,ddsYi(t,s)=Di(X(t,s),Y(t,s),t−s)),Y(t,0)=w.\begin{cases} \frac{d}{ds}X^i(t,s)=-B^i(X(t,s),Y(t,s),t-s)),\qquad X(t,0)=x,\\ \frac{d}{ds}Y^i(t,s)=D^i(X(t,s),Y(t,s),t-s)),\qquad Y(t,0)=w. \end{cases}

Define

h(s)=f(Y(t,s);X(t,s),t−s)eN(w;x,s)h(s)=f(Y(t,s);X(t,s),t-s)e^{N(w;x,s)}

for all s∈[0,t]s\in[0,t], where

N(w;x,s):=∫0s(∂Di∂wi−∂Bi∂xi)(X(t,r);Y(t,r),t−r) dr.N(w;x,s):=\int_0^s\left(\frac{\partial D^i}{\partial w^i}-\frac{\partial B^i}{\partial x^i}\right)(X(t,r);Y(t,r),t-r)\,dr.

Then clearly h(0)=f(w;x,t)h(0)=f(w;x,t) and

h(t)=f(Y(t,t);X(t,t),0)eN(w;x,t).h(t)=f(Y(t,t);X(t,t),0)e^{N(w;x,t)}.

It is clear that

ddsh(s)=eN∂f∂widdsYi(t,s)+eN∂f∂xiddsXi(t,s)−eN∂∂tf+eNf∂Di∂wi−eNf∂Bi∂xi=eN{∂(fDi)∂wi−∂(Bif)∂xi−∂∂tf}=0.\begin{aligned} \frac{\mathrm{d}}{\mathrm{d}s}h(s) &=e^{N}\frac{\partial f}{\partial w^{i}}\frac{\mathrm{d}}{\mathrm{d}s}Y^{i}(t,s)+e^{N}\frac{\partial f}{\partial x^{i}}\frac{\mathrm{d}}{\mathrm{d}s}X^{i}(t,s)\\ &\quad-e^{N}\frac{\partial}{\partial t}f+e^{N}f\frac{\partial D^{i}}{\partial w^{i}}-e^{N}f\frac{\partial B^{i}}{\partial x^{i}}\\ &=e^{N}\left\{\frac{\partial(fD^{i})}{\partial w^{i}}-\frac{\partial(B^{i}f)}{\partial x^{i}}-\frac{\partial}{\partial t}f\right\}\\ &=0. \end{aligned}

Integrating with respect to the variable ss over [0,t][0,t], we may conclude that

f(w;x,t)=f(Y(t,t);X(t,t),0)exp⁡[∫0t(∂Di∂wi−∂Bi∂xi)(X(t,s);Y(t,s),t−s) ds].f(w;x,t)=f(Y(t,t);X(t,t),0)\exp\left[\int_{0}^{t}\left(\frac{\partial D^{i}}{\partial w^{i}}-\frac{\partial B^{i}}{\partial x^{i}}\right)(X(t,s);Y(t,s),t-s)\,\mathrm{d}s\right].

The conservation of positivity property 1) follows immediately from this representation. To show 2) we consider the function u(x,t)=∫R3f(w;x,t) dwu(x,t)=\int_{\mathbb{R}^{3}}f(w;x,t)\,\mathrm{d}w for x∈R3x\in\mathbb{R}^{3} and t≥0t\ge0. If u(x,t)=1u(x,t)=1 for all xx and t>0t>0, then by integrating (18) over R3\mathbb{R}^{3} we obtain (19). Conversely, if (19) holds, by integrating (18) over R3\mathbb{R}^{3} then ∂∂tu(x,t)=0\frac{\partial}{\partial t}u(x,t)=0, which yields that u(x,t)=1u(x,t)=1.

Weakly homogeneous and weakly isotropic flows

It has been pointed out in Secs. 2 and 3, the coefficients appearing in the PDF PDE are determined by the conditional mean function μ(x,y,w,t)\mu(x,y,w,t) defined in (1). The significance of this statistical quantity of a turbulent flow has been demonstrated in the derivation of the PDF PDE. By definition the conditional mean describes the deviation from the local isotropicity of the turbulent flow, and therefore this statistical characteristic can be used in the classification of turbulent flows. In this section, we define weak homogeneous and weak isotropic turbulent flows, then study the PDF equation for these turbulent flows.

The homogeneity and the isotropicity can be defined in general for random fields indexed by a space variable x∈Rdx\in\mathbb{R}^{d}, which has been introduced into the study of turbulence by G. I. Taylor [19]. The local homogeneous and local isotropic flows were introduced by Kolmogorov for formulating K41 theory (and its improved version K62 theory). According to Kolmogorov [9, 8], a random field {Z(x,t)}\{Z(x,t)\} is locally homogeneous if the conditional distribution of Z(y,t)−Z(x,s)Z(y,t)-Z(x,s) given Z(x,s)=zZ(x,s)=z is independent of (z,x,s)(z,x,s) for any t≥st\ge s, and further it is locally isotropic if the conditional distribution of Z(y,t)−Z(x,s)Z(y,t)-Z(x,s) are invariant under reflections and rotations. The usage of conditional mean function makes it possible to generalize these concepts to their weak versions.

We may simplify the PDF PDE for turbulent flows when the vorticity random field W(x,t)W(x,t) is weakly homogeneous in the sense that for every pair (y,x)(y,x) and t>0t>0, the conditional mean of Wi(y,t)−Wi(x,t)W^{i}(y,t)-W^{i}(x,t) given W(x,t)=wW(x,t)=w depends only on y−xy-x, ww and t>0t>0 but independent of xx, so that we may write

μi(x,y,w,t)=βi(y−x,w,t),\mu^{i}(x,y,w,t)=\beta^{i}(y-x,w,t),

where βi(0,w,t)=0\beta^{i}(0,w,t)=0. Then

aki(x,w,t)=∂βi∂xk(0,w,t),a^{i}_{k}(x,w,t)=\frac{\partial\beta^{i}}{\partial x^{k}}(0,w,t),

and

bi(x,w,t)=Δβi(0,w,t),b^{i}(x,w,t)=\Delta\beta^{i}(0,w,t),

which are still denoted by aki(w,t)a^{i}_{k}(w,t) and bi(w,t)b^{i}(w,t), though they are independent of xx. Similarly

Bi(w,t)=∫R314π∣y∣εijk∂βk∂yj(y,w,t) dy,B^{i}(w,t)=\int_{\mathbb{R}^{3}}\frac{1}{4\pi|y|}\varepsilon^{ijk}\frac{\partial\beta^{k}}{\partial y^{j}}(y,w,t)\,dy,

and

Di(w,t)=wl∫R3yl4π∣y∣3εijk∂βk∂yj(y,w,t) dy,D^{i}(w,t)=w^{l}\int_{\mathbb{R}^{3}}\frac{y^{l}}{4\pi|y|^{3}}\varepsilon^{ijk}\frac{\partial\beta^{k}}{\partial y^{j}}(y,w,t)\,dy,

which again are functions of (w,t)(w,t) only. Therefore the PDF PDE can be simplified to be the following mixed type PDE

(∂∂t+Bi∂∂xi−νΔ)f=ν∂∂wi(aki∂f∂xk)+∂∂wi(fAi),\left(\frac{\partial}{\partial t}+B^{i}\frac{\partial}{\partial x^{i}}-\nu\Delta\right)f =\nu\frac{\partial}{\partial w^{i}}\left(a^{i}_{k}\frac{\partial f}{\partial x^{k}}\right) +\frac{\partial}{\partial w^{i}}(fA^{i}),

where Ai=Di−2νbiA^{i}=D^{i}-2\nu b^{i} and i=1,2,3i=1,2,3.

We say that the vorticity WW is weakly isotropic, if

E[∇W(x,t)∣W(x,t)=w]=0\mathbb{E}\left[\nabla W(x,t)\mid W(x,t)=w\right]=0

for all x,wx,w and tt. By definition, WW is locally isotropic in Kolmogorov’s sense, then it is weakly isotropic.

If WW is weakly homogeneous and weakly isotropic, then a=0a=0, BB and AA depend on (w,t)(w,t) only, so for this case, the PDF PDE becomes a parabolic-transport equation

(∂∂t+Bi∂∂xi−νΔ)f=∂∂wi(fAi).(20)\left(\frac{\partial}{\partial t}+B^{i}\frac{\partial}{\partial x^{i}}-\nu\Delta\right)f =\frac{\partial}{\partial w^{i}}(fA^{i}). \tag*{(20)}

After we have deduced the PDE (20), there is no need to assume that BB and AA are independent of xx. Therefore we may consider the following PDE

(∂∂t+Bi∂∂xi+∂Bi∂xi−νΔ)f=∂∂wi(fAi).(21)\left(\frac{\partial}{\partial t}+B^{i}\frac{\partial}{\partial x^{i}}+\frac{\partial B^{i}}{\partial x^{i}}-\nu\Delta\right)f =\frac{\partial}{\partial w^{i}}(fA^{i}). \tag*{(21)}

where both BB and AA are functions of three variables xx, ww and tt.

The following theorem provides the foundation for modeling weakly isotropic turbulent flows based on the vorticity PDF.

Theorem 6. Consider the parabolic-transport differential equation (21), where AA and BB are considered as given. Assume that A(x,w,t)A(x,w,t) and B(x,w,t)B(x,w,t) are Lipschitz continuous in space variable (x,w)(x,w) uniformly in tt. Suppose f(w;x,t)f(w;x,t) is a solution to (21) such that fAfA is continuous and decays to zero sufficiently fast as ∣w∣→∞|w|\to\infty.

  1. If f(w;x,0)≥0f(w;x,0)\ge0 for all xx and ww, then f(w;x,t)≥0f(w;x,t)\ge0 for all t≥0t\ge0, xx and ww.

  1. If f(w;x,0)f(w;x,0) is a PDF for every xx, i.e. it is non-negative and ∫R3f(w;x,0) dw=1\int_{\mathbb{R}^{3}}f(w;x,0)\,dw=1 for every xx, then so are f(w;x,t)f(w;x,t) for all xx and t>0t>0 if and only if

∂∂xi∫R3Bi(x,w,t)f(w;x,t) dw=0(22)\frac{\partial}{\partial x^{i}}\int_{\mathbb{R}^{3}}B^{i}(x,w,t)f(w;x,t)\,dw=0 \tag*{(22)}

for all xx and t>0t>0.

  1. Suppose in addition BB is independent of ww and ∂Bi∂xi=0\frac{\partial B^{i}}{\partial x^{i}}=0, then if f(w;x,0)f(w;x,0) is a PDF for every xx, f(w;x,t)f(w;x,t) is also a PDF for every xx and for every $t>0.

Proof. The proof is rather similar to that of Theorem 5, so we only provide an outline. Let (Mti)t≥0(M_t^i)_{t\geq0} (i=1,2,3i=1,2,3) be a three dimensional standard Brownian motion on a probability space (Ω,F,P)(\Omega,\mathcal{F},\mathbb{P}). For every fixed t>0t>0 we run the following SDE system:

dXi(t,s)=2ν dMsi−Bi(X(t,s),Y(t,s),t−s)ds,X(t,0)=x,\mathrm{d}X^i(t,s)=\sqrt{2\nu}\,\mathrm{d}M_s^i-B^i(X(t,s),Y(t,s),t-s)\mathrm{d}s,\qquad X(t,0)=x,
dYi(t,s)=Ai(X(t,s),Y(t,s),t−s)ds,Y(t,0)=w,\mathrm{d}Y^i(t,s)=A^i(X(t,s),Y(t,s),t-s)\mathrm{d}s,\qquad Y(t,0)=w,

which has a unique solution running up to tt. Apply Itô’s formula to

Hs=f(Y(t,s);X(t,s),t−s)eP(s),H_s=f(Y(t,s);X(t,s),t-s)e^{P(s)},

where

P(s)=∫0s∂Ai∂wi(X(t,r),Y(t,r),t−r)dr.P(s)=\int_0^s\frac{\partial A^i}{\partial w^i}(X(t,r),Y(t,r),t-r)\mathrm{d}r.

Then

dHs=eP(∂f∂widYsi+∂f∂xidXsi+f∂Ai∂wi−∂f∂sds+νΔfds)=2νeP∂f∂xidMsi.\begin{aligned} \mathrm{d}H_s=e^P\left(\frac{\partial f}{\partial w^i}\mathrm{d}Y_s^i+\frac{\partial f}{\partial x^i}\mathrm{d}X_s^i+f\frac{\partial A^i}{\partial w^i}-\frac{\partial f}{\partial s}\mathrm{d}s+\nu\Delta f\mathrm{d}s\right) \\ &=\sqrt{2\nu}e^P\frac{\partial f}{\partial x^i}\mathrm{d}M_s^i. \end{aligned}

so that

H0=Ht−2νeP∂f∂xidMsi.H_0=H_t-\sqrt{2\nu}e^P\frac{\partial f}{\partial x^i}\mathrm{d}M_s^i.

By taking expectation, one may deduce that the solution can be expressed as

f(w;x,t)=E[f(Y(t,t);X(t,t),0)exp⁡(∫0t∂Ai∂wi(X(t,s),Y(t,s),t−s)ds)].f(w;x,t)=\mathbb{E}\left[f(Y(t,t);X(t,t),0)\exp\left(\int_0^t\frac{\partial A^i}{\partial w^i}(X(t,s),Y(t,s),t-s)\mathrm{d}s\right)\right].

Statements 1) - 3) can be easily shown using this expression and the argument in the proof of Theorem 5. If BB is independent of ww, then, by integrating (20),

(∂∂t+Bi∂∂xi−νΔ)u(x,t)=0,\left(\frac{\partial}{\partial t}+B^i\frac{\partial}{\partial x^i}-\nu\Delta\right)u(x,t)=0,

where u(x,t)=∫R3f(w;x,t)dwu(x,t)=\int_{\mathbb{R}^3}f(w;x,t)\mathrm{d}w. Thus 3) follows from the uniqueness of the solution of the previous parabolic equation.

Modeling PDF of weakly isotropic flows

In this and the next sections, several simple models based on our PDF equation for modeling vorticity distributions in turbulence are discussed. Clearly, the most straightforward way for modeling the distribution of vorticity is to assign the conditional mean function μ(x,y,w,t)\mu(x,y,w,t). The other parameters in the PDF PDE may be determined accordingly.

For practical reasons which will be clarified in our computations below, the simplest yet not trivial model for μ(x,y,w,t)\mu(x,y,w,t) should be a function of ∣y−x∣2|y-x|^2 only and μ\mu decays sufficiently fast at infinity. For such a model, the parameters BiB^i and DiD^i appearing in the PDF PDE (8) vanish identically. The other parameters akia_k^i and bjkib_{jk}^i are determined by the asymptotic condition as ∣y−x∣→0|y-x|\to0. For this reason, we assign

μi(x,y,w,t)={−C6ν∣y−x∣2,if ∣x−y∣<δ,h(∣y−x∣2),if ∣x−y∣≥δ,(23)\mu^i(x,y,w,t)= \begin{cases} -\dfrac{C}{6\nu}|y-x|^2, & \text{if } |x-y|<\delta,\\ h(|y-x|^2), & \text{if } |x-y|\geq\delta, \end{cases} \tag*{(23)}

for i=1,2,3i=1,2,3, where δ>0\delta>0 and hh is a C1C^1-function such that hh and its derivative decay sufficiently fast at infinity. C≥0C\geq0 is a model parameter which should be reflecting certain physical properties of the underlying flows. The other parameters can be read out from this model easily: aki=0a_k^i=0, bjki=−δjkC6νb_{jk}^i=-\delta_{jk}\frac{C}{6\nu} and therefore bi=−C2νb^i=-\frac{C}{2\nu}. With this model of μ\mu, the PDF PDE is simplified to be the following parabolic-transport equation:

(∂∂t−νΔ)f=C∑i∂f∂wi.(24)\left(\frac{\partial}{\partial t}-\nu\Delta\right)f=C\sum_i\frac{\partial f}{\partial w^i}. \tag*{(24)}

The solution to (24) has a nice probabilistic representation which may be read out from the proof of Theorem 6. In fact

f(w;x,t)=E[f(Y(t,t);X(t,t),0)](25)f(w;x,t)=\mathbb{E}\left[f(Y(t,t);X(t,t),0)\right] \tag*{(25)}

where f(w;x,0)f(w;x,0) is the PDF of the initial vorticity, and XX and YY are solutions to the SDE:

{dXi(t,s)=2ν dMsi,X(t,0)=x,dYi(t,s)=C ds,Y(t,0)=w,\begin{cases} dX^i(t,s)=\sqrt{2\nu}\,dM_s^i,\quad X(t,0)=x,\\ dY^i(t,s)=C\,ds,\quad Y(t,0)=w, \end{cases}

where MM is a standard 3D Brownian motion. The solutions when s=ts=t are given by

Yi(t,t)=wi+CtY^i(t,t)=w^i+Ct

and

X(t,t)=x+2νMt.X(t,t)=x+\sqrt{2\nu}M_t.

Substituting these into (25) we obtain

f(w;x,t)=E[f(w+Ct;x+2νMt,0)]=E[f(w+Ct;x+2νtξ,0)],(26)\begin{aligned} f(w;x,t)&=\mathbb{E}\left[f(w+Ct;x+\sqrt{2\nu}M_t,0)\right]\\ &=\mathbb{E}\left[f(w+Ct;x+\sqrt{2\nu t}\xi,0)\right], \tag*{(26)} \end{aligned}

where ξ\xi is a random vector with the standard 3D normal distribution N(0,I3)N(0,I_3). Clearly the representation (26) may also be written in terms of Gaussian density

f(w;x,t)=1(4πνt)3/2∫R3f(w+Ct,z,0)e−∣z−x∣24νt dz=1(2π)3/2∫R3f(w+Ct,x+2νtz,0)e−∣z∣22 dz.(27)\begin{aligned} f(w;x,t)&=\frac{1}{(4\pi\nu t)^{3/2}}\int_{\mathbb{R}^3}f(w+Ct,z,0)e^{-\frac{|z-x|^2}{4\nu t}}\,dz\\ &=\frac{1}{(2\pi)^{3/2}}\int_{\mathbb{R}^3}f(w+Ct,x+\sqrt{2\nu t}z,0)e^{-\frac{|z|^2}{2}}\,dz. \tag*{(27)} \end{aligned}

Both representations (26) and (27) may be used to evaluate f(w;x,t)f(w;x,t) by using for example Monte-Carlo scheme by sampling Gaussian random variables.

The initial PDF f(w;x,0)f(w;x,0) is determined by the initial vorticity distribution. For a turbulent flow, the initial vorticity may be written as a sum:

W(x,0)=ω0(x)+ε(x),W(x,0)=\omega_0(x)+\varepsilon(x),

where ω0(x)\omega_0(x) is the initial mean vorticity at the location xx, and ε(x)\varepsilon(x) represents a small random perturbation. It is reasonable, therefore, to assume that ε(x)\varepsilon(x) has a normal distribution N(0,σ(x)2I3)N(0,\sigma(x)^2I_3), where the variance σ(x)\sigma(x) may or may not depend on the location xx. However, if ε(x)\varepsilon(x) is allowed to depend on the location xx, then the noise ε(x)\varepsilon(x) has to satisfy the divergence-free condition as well, a technical issue we will not address here in detail. It is reasonable to assume that the initial vorticity mean ω0\omega_0 is distributed in a small region of the space. In particular ω0\omega_0 decays to zero sufficiently fast near infinity. Under this assumption,

f(w;x,0)=1(2πσ(x)2)3/2exp⁡[−∣w−ω0(x)∣22σ(x)2].(28)f(w;x,0)=\frac{1}{(2\pi\sigma(x)^2)^{3/2}}\exp\left[-\frac{|w-\omega_0(x)|^2}{2\sigma(x)^2}\right]. \tag*{(28)}

By substituting this into the representation (26), we obtain

f(w;x,t)=E[1(2πσ(x+2νtξ)2)3/2e−∣w+Ct−ω0(x+2νtξ)∣22σ(x+2νtξ)2].(29)f(w;x,t)=\mathbb{E}\left[\frac{1}{\left(2\pi\sigma\left(x+\sqrt{2\nu t}\xi\right)^2\right)^{3/2}}e^{-\frac{|w+Ct-\omega_0\left(x+\sqrt{2\nu t}\xi\right)|^2}{2\sigma\left(x+\sqrt{2\nu t}\xi\right)^2}}\right]. \tag*{(29)}

It is easy to see that f(w;x,t)f(w;x,t) is no longer Gaussian except for special cases which we would like to discuss below.

If ω0\omega_0 vanishes and if σ\sigma is a positive small constant, the initial distribution is homogeneous and equation (29) leads to a simple expression

f(w;x,t)=1(2πσ2)3/2exp⁡(−∣w+Ct∣22σ2),f(w;x,t)=\frac{1}{(2\pi\sigma^2)^{3/2}}\exp\left(-\frac{|w+Ct|^2}{2\sigma^2}\right),

which is independent of the location xx and remains a Gaussian density. The interesting feature about this model is that the variance stays as the constant σ2\sigma^2 but new mean vorticity −Ct-Ct is created evenly after duration tt. This is the case of a turbulence with a small constant random perturbation. For turbulent flows observed in nature, the initial vorticity mean ω0\omega_0 does exist and does not vanish, and the random noise ε(x)\varepsilon(x), for simplicity, may be modeled by a Gaussian random variable independent of xx. Therefore when σ>0\sigma>0 is a small constant, we have

f(w;x,t)=1(2πσ2)3/2E[e−∣w+Ct−ω0(x+2νtξ)∣22σ2],(30)f(w;x,t)=\frac{1}{(2\pi\sigma^2)^{3/2}}\mathbb{E}\left[e^{-\frac{|w+Ct-\omega_0\left(x+\sqrt{2\nu t}\xi\right)|^2}{2\sigma^2}}\right], \tag*{(30)}

where ξ∼N(0,I3)\xi\sim N(0,I_3).

For this case the mean vorticity ω(x,t)\omega(x,t) at (x,t)(x,t) can be evaluated. Indeed

ω(x,t)=∫R3w(2πσ2)3/2E[e−∣w+Ct−ω0(x+2νtξ)∣22σ2] dw=E[∫R3w(2πσ2)3/2e−∣w+Ct−ω0(x+2νtξ)∣22σ2 dw]=E[ω0(x+2νtξ)]−Ct,\begin{aligned} \omega(x,t)=\int_{\mathbb{R}^3}\frac{w}{(2\pi\sigma^2)^{3/2}}\mathbb{E}\left[e^{-\frac{|w+Ct-\omega_0\left(x+\sqrt{2\nu t}\xi\right)|^2}{2\sigma^2}}\right]\,dw \\ &=\mathbb{E}\left[\int_{\mathbb{R}^3}\frac{w}{(2\pi\sigma^2)^{3/2}}e^{-\frac{|w+Ct-\omega_0\left(x+\sqrt{2\nu t}\xi\right)|^2}{2\sigma^2}}\,dw\right] \\ &=\mathbb{E}\left[\omega_0\left(x+\sqrt{2\nu t}\xi\right)\right]-Ct, \end{aligned}

and therefore

ωi(x,t)=∫R3ω0i(y)(4πνt)3/2e−∣y−x∣24νt dy−Ct.(31)\omega^i(x,t)=\int_{\mathbb{R}^3}\frac{\omega_0^i(y)}{(4\pi\nu t)^{3/2}}e^{-\frac{|y-x|^2}{4\nu t}}\,dy-Ct. \tag*{(31)}

This equality shows that the vorticity mean ω(x,t)\omega(x,t) under this simple model is independent of the noise parameter σ2\sigma^2, and ω(x,t)\omega(x,t) evolves according to the heat type equation

∂∂tω(x,t)−C=νΔω(x,t),ω(x,0)=ω0(x),\frac{\partial}{\partial t}\omega(x,t)-C=\nu\Delta\omega(x,t),\quad\omega(x,0)=\omega_0(x),

which is a rather crude approximation to the mean vorticity equation.

Heat flow method

In this section we propose another model for the PDF of the vorticity, based on the heat flow method, in which the conditional mean function μ\mu is generated by a random field R(x,τ)R(x,\tau). The random field R(x,τ)R(x,\tau) evolves according to the heat flow

(∂∂τ−νΔ)Ri(x,τ)=0,(32)\left(\frac{\partial}{\partial\tau}-\nu\Delta\right)R^i(x,\tau)=0, \tag*{(32)}

where R(x,τ)=(R1(x,τ),R2(x,τ),R3(x,τ))R(x,\tau)=(R^1(x,\tau),R^2(x,\tau),R^3(x,\tau)). The initial value R(x,0)=ξ(x)=(ξi(x))R(x,0)=\xi(x)=(\xi^i(x)) is a centered Gaussian noise white in space in the sense that the co-variance of ξi\xi^i at two locations uu and vv in R3\mathbb{R}^3 is given by

E[ξi(u)ξi(v)]=δ(u−v).(33)\mathbb{E}\left[\xi^i(u)\xi^i(v)\right]=\delta(u-v). \tag*{(33)}

Also, we assume that ξi\xi^i's are i.i.d. random variables. For simplicity, we assume that RiR^i is centered here, but our argument can definitely be generalized to the case when it is non-centered.

The solution to equation (32) is given by

Ri(x,τ)=∫R31(4πντ)32e−∣x−y∣24ντξi(y) dy,(34)R^i(x,\tau)=\int_{\mathbb{R}^3}\frac{1}{(4\pi\nu\tau)^{\frac{3}{2}}}e^{-\frac{|x-y|^2}{4\nu\tau}}\xi^i(y)\,dy, \tag*{(34)}

so that R(x,τ)R(x,\tau) is a centered Gaussian random field with its co-variance

στ(x,y)=∫R3∫R31(4πντ)3e−∣x−u∣2+∣y−v∣24ντE[ξi(u)ξi(v)] du dv=∫R3∫R31(4πντ)3e−∣x−u∣2+∣y−v∣24ντδ(u−v) du dv,=e−∣x−y∣28ντ∫R31(4πντ)3e−∣u−x+y2∣22ντ du=18(2πντ)32e−∣x−y∣28ντ.\begin{aligned} \sigma_\tau(x,y)&=\int_{\mathbb{R}^3}\int_{\mathbb{R}^3}\frac{1}{(4\pi\nu\tau)^3}e^{-\frac{|x-u|^2+|y-v|^2}{4\nu\tau}}\mathbb{E}\left[\xi^i(u)\xi^i(v)\right]\,du\,dv\\ &=\int_{\mathbb{R}^3}\int_{\mathbb{R}^3}\frac{1}{(4\pi\nu\tau)^3}e^{-\frac{|x-u|^2+|y-v|^2}{4\nu\tau}}\delta(u-v)\,du\,dv,\\ &=e^{-\frac{|x-y|^2}{8\nu\tau}}\int_{\mathbb{R}^3}\frac{1}{(4\pi\nu\tau)^3}e^{-\frac{\left|u-\frac{x+y}{2}\right|^2}{2\nu\tau}}\,du\\ &=\frac{1}{8(2\pi\nu\tau)^{\frac{3}{2}}}e^{-\frac{|x-y|^2}{8\nu\tau}}. \end{aligned}

It follows that the conditional distribution of R(y,τ)R(y,\tau) given R(x,τ)=wR(x,\tau)=w has a normal distribution with mean e−∣x−y∣28ντwe^{-\frac{|x-y|^2}{8\nu\tau}}w and co-variance matrix

18(2πντ)32(1−e−∣x−y∣24ντ)I3.\frac{1}{8(2\pi\nu\tau)^{\frac{3}{2}}}\left(1-e^{-\frac{|x-y|^2}{4\nu\tau}}\right)I_3.

Therefore the conditional mean of R(y,τ)−R(x,τ)R(y,\tau)-R(x,\tau) given R(x,τ)=wR(x,\tau)=w can be easily found to be

−(1−e−∣x−y∣28ντ)w,-\left(1-e^{-\frac{|x-y|^2}{8\nu\tau}}\right)w,

which will be our μ(x,y,w,t)\mu(x,y,w,t) with τ=φ(t)\tau=\varphi(t) a reparametrization as part of the model. For simplicity let us consider the power law model, that is,

τ=λ1(t+λ2)α,(35)\tau=\lambda_1(t+\lambda_2)^\alpha, \tag*{(35)}

where λ1\lambda_1 and λ2\lambda_2 are two positive numbers which then become our model parameters. Since

μ(x,y,w,t)=−∣x−y∣28ντw+o(∣x−y∣2)w,\mu(x,y,w,t)=-\frac{|x-y|^2}{8\nu\tau}w+o(|x-y|^2)w,

therefore aki=0a_k^i=0 and bjki=−δjk18νtwib_{jk}^i=-\delta_{jk}\frac{1}{8\nu t}w^i. In particular,

bi(w,t)=−3wi8νλ1(t+λ2)α.(36)b^i(w,t)=-\frac{3w^i}{8\nu\lambda_1(t+\lambda_2)^\alpha}. \tag*{(36)}

for i=1,2,3i=1,2,3. Since the conditional mean function μ\mu depends only on ∣x−y∣2|x-y|^2 and decays exponentially fast at infinity, Bi=0B^i=0 and Di=0D^i=0, which implies that

Ai(w,t)=3wi4λ1(t+λ2)α.(37)A^i(w,t)=\frac{3w^i}{4\lambda_1(t+\lambda_2)^\alpha}. \tag*{(37)}

Thus the PDF PDE with these parameters is reduced to the simple parabolic-transport equation

(∂∂t−νΔx)f=∂∂wi(Aif).\left(\frac{\partial}{\partial t}-\nu\Delta_x\right)f=\frac{\partial}{\partial w^i}\left(A^i f\right).

The divergence of AA is given by

∂Ai∂wi(x,w,t)=94λ1(t+λ2)α.\frac{\partial A^i}{\partial w^i}(x,w,t)=\frac{9}{4\lambda_1(t+\lambda_2)^\alpha}.

If α≠1\alpha\ne1 but α>0\alpha>0, by using the Feynman-Kac formula we have

f(w;x,t)=θ(α,t)3E[f(Y(t,t),X(t,t),0)],(38)f(w;x,t)=\theta(\alpha,t)^{3\mathbb{E}}[f(Y(t,t),X(t,t),0)], \tag*{(38)}

where we have introduced

θ(α,t)=exp⁡[3((t+λ2)1−α−λ21−α)4λ1(1−α)](39)\theta(\alpha,t)=\exp\left[\frac{3\left((t+\lambda_2)^{1-\alpha}-\lambda_2^{1-\alpha}\right)}{4\lambda_1(1-\alpha)}\right] \tag*{(39)}

for simplicity, and XX and YY are solutions to SDE

{dXi(t,s)=2ν dMsi,X(t,0)=x,dYi(t,s)=3Yi(t,s)4λ1(t−s+λ2)α ds,Y(t,0)=w.\begin{cases} dX^i(t,s)=\sqrt{2\nu}\,dM_s^i, & X(t,0)=x,\\ dY^i(t,s)=\dfrac{3Y^i(t,s)}{4\lambda_1(t-s+\lambda_2)^\alpha}\,ds, & Y(t,0)=w. \end{cases}

If α=1\alpha=1 then

f(w;x,t)=λ(t)3E[f(Y(t,t),X(t,t),0)],(40)f(w;x,t)=\lambda(t)^{3\mathbb{E}}[f(Y(t,t),X(t,t),0)], \tag*{(40)}

where

λ(t)=(t+λ2λ2)34λ1.(41)\lambda(t)=\left(\frac{t+\lambda_2}{\lambda_2}\right)^{\frac{3}{4\lambda_1}}. \tag*{(41)}

The previous SDEs have explicit solutions:

X(t,t)=x+2νMtX(t,t)=x+\sqrt{2\nu}M_t

and

Y(t,t)=wθ(α,t)Y(t,t)=w\theta(\alpha,t)

if α≠1\alpha\ne1, and

Y(t,t)=λ(t)wY(t,t)=\lambda(t)w

if α=1\alpha=1.

Therefore, if α>0\alpha> 0 and α≠1\alpha\ne1, by plugging these explicit solutions XX and YY into (38) we have

f(w;x,t)=θ(α,t)3E[f(wθ(α,t),x+2νMt,0)]=θ(α,t)3E[f(wθ(α,t),x+2νtξ,0)](42)f(w;x,t)=\theta(\alpha,t)^3\mathbb{E}\left[f\left(w\theta(\alpha,t),x+\sqrt{2\nu}M_t,0\right)\right] =\theta(\alpha,t)^3\mathbb{E}\left[f\left(w\theta(\alpha,t),x+\sqrt{2\nu t}\xi,0\right)\right] \tag*{(42)}

where ξ∼N(0,I3)\xi\sim N(0,I_3).

As in the previous section, the initial vorticity is assumed to be of the form

W(x,0)=ω(x)+ε(x),W(x,0)=\boldsymbol{\omega}(x)+\boldsymbol{\varepsilon}(x),

where ε(x)\boldsymbol{\varepsilon}(x) has a normal distribution N(0,σ2)N(0,\sigma^2) and ω(x)\boldsymbol{\omega}(x) is the mean vorticity at xx, so that

f(w;x,0)=1(2πσ2)3/2exp⁡[−∣w−ω(x)∣22σ2].f(w;x,0)=\frac{1}{(2\pi\sigma^2)^{3/2}}\exp\left[-\frac{\left|w-\boldsymbol{\omega}(x)\right|^2}{2\sigma^2}\right].

and therefore, according to (42),

f(w;x,t)=θ(α,t)3E[1(2πσ(x+2νtξ)2)3/2e−∣wθ(α,t)−ω(x+2νtξ)∣22σ(x+2νtξ)2],f(w;x,t)=\theta(\alpha,t)^3\mathbb{E}\left[\frac{1}{\left(2\pi\sigma(x+\sqrt{2\nu t}\xi)^2\right)^{3/2}}e^{-\frac{\left|w\theta(\alpha,t)-\boldsymbol{\omega}(x+\sqrt{2\nu t}\xi)\right|^2}{2\sigma(x+\sqrt{2\nu t}\xi)^2}}\right],

where ξ∼N(0,I3)\xi\sim N(0,I_3). Hence f(w;x,t)f(w;x,t) is no longer Gaussian density unless σ(x)\sigma(x) is independent of xx and ω=0\boldsymbol{\omega}=0 identically.

Let us discuss a special case where ω=0\boldsymbol{\omega}=0 identically and σ>0\sigma>0 is constant. Then

f(w;x,t)=θ(α,t)3(2πσ2)3/2exp⁡[−∣wθ(α,t)∣22σ2]f(w;x,t)=\frac{\theta(\alpha,t)^3}{(2\pi\sigma^2)^{3/2}}\exp\left[-\frac{\left|w\theta(\alpha,t)\right|^2}{2\sigma^2}\right]

is a Gaussian density with mean 0 and variance ρ2I3\rho^2 I_3, where

ρ2(x,t)=1θ(α,t)σ2.(43)\rho^2(x,t)=\frac{1}{\sqrt{\theta(\alpha,t)}}\sigma^2. \tag*{(43)}

If σ>0\sigma>0 is a constant and ω\boldsymbol{\omega} does not vanish identically, then f(w;x,t)f(w;x,t) is not Gaussian, but has a nice representation:

f(w;x,t)=θ(α,t)3(2πσ2)3/2E[e−∣wθ(α,t)−ω(x+2νtξ)∣22σ2],(44)f(w;x,t)=\frac{\theta(\alpha,t)^3}{(2\pi\sigma^2)^{3/2}}\mathbb{E}\left[e^{-\frac{\left|w\theta(\alpha,t)-\boldsymbol{\omega}(x+\sqrt{2\nu t}\xi)\right|^2}{2\sigma^2}}\right], \tag*{(44)}

where ξ∼N(0,I3)\xi\sim N(0,I_3), and θ(α,t)\theta(\alpha,t) is defined by (39).

A similar discussion applies to the model where α=1\alpha=1, which is certainly interesting too. For this model the PDF f(w;x,t)f(w;x,t) is given by

f(w;x,t)=λ(t)3E[f(wλ(t),x+2νtξ,0)](45)f(w;x,t)=\lambda(t)^3\mathbb{E}\left[f\left(w\lambda(t),x+\sqrt{2\nu t}\xi,0\right)\right] \tag*{(45)}

where ξ∼N(0,I3)\xi\sim N(0,I_3) and λ(t)\lambda(t) is defined by (41).

Suppose again the initial vorticity

W(x,0)=ω(x)+ε(x),W(x,0)=\boldsymbol{\omega}(x)+\boldsymbol{\varepsilon}(x),

where ε(x)\varepsilon(x) has a normal distribution N(0,σ2(x))N(0,\sigma^2(x)) and ω(x)\omega(x) is the mean. Then

f(w;x,t)=E[λ(t)3(2πσ(x+2νtξ)2)3/2e−∣wλ(t)−ω(x+2νtξ)∣22σ(x+2νtξ)2](46)f(w;x,t)=\mathbb{E}\left[\frac{\lambda(t)^3}{\left(2\pi\sigma\left(x+\sqrt{2\nu t}\xi\right)^2\right)^{3/2}}e^{-\frac{\left\lvert w\lambda(t)-\omega\left(x+\sqrt{2\nu t}\xi\right)\right\rvert^2}{2\sigma\left(x+\sqrt{2\nu t}\xi\right)^2}}\right] \tag*{(46)}

which is not Gaussian except for the following special case. If σ(x)=σ>0\sigma(x)=\sigma>0 is a constant and ω(x)=0\omega(x)=0, then

f(w;x,t)=λ(t)3(2πσ2)3/2e−∣wλ(t)∣22σ2f(w;x,t)=\frac{\lambda(t)^3}{\left(2\pi\sigma^2\right)^{3/2}}e^{-\frac{\left\lvert w\lambda(t)\right\rvert^2}{2\sigma^2}}

is Gaussian with mean zero and variance ρ2I3\rho^2 I_3, where

ρ2(x,t)=σ2(λ2t+λ2)32λ1.\rho^2(x,t)=\sigma^2\left(\frac{\lambda_2}{t+\lambda_2}\right)^{\frac{3}{2\lambda_1}}.

The most interesting case for the purpose of modeling turbulent flows is the model where σ>0\sigma>0 is a small parameter and ω\omega is not zero, so that

f(w;x,t)=λ(t)3(2πσ2)3/2E[e−∣wλ(t)−ω(x+2νtξ)∣22σ2]f(w;x,t)=\frac{\lambda(t)^3}{\left(2\pi\sigma^2\right)^{3/2}}\mathbb{E}\left[e^{-\frac{\left\lvert w\lambda(t)-\omega\left(x+\sqrt{2\nu t}\xi\right)\right\rvert^2}{2\sigma^2}}\right]

where ξ∼N(0,I3)\xi\sim N(0,I_3). Although it is no longer Gaussian, some of its features can be extracted by doing Monte-Carlo simulations.

We may conclude that this model provides some nice features of the propagation of the vorticity which may be helpful for the understanding of the energy dissipation in turbulence, and we will explore this in a separate work.

Bibliography

References

  1. [1]G. K. Batchelor. The Theory of Homogeneous Turbulence. Cambridge University Press, 1953.
  2. [3]G.-H. Cottet and P. D. Koumoutsakos. Vortex Methods: Theory and Practice. Cambridge University Press, 2000.
  3. [4]P. A. Davidson. Turbulence: An Introduction for Scientists and Engineers. Oxford University Press, 1995.
  4. [6]E. Hopf. Statistical hydromechanics and functional calculus. Journal of Rational Mechanics and Analysis, 1:87–123, 1952.DOI
  5. [8]A. N. Kolmogorov. The local structure of turbulence in incompressible viscous fluid for very large Reynolds numbers. Proceedings of the Royal Society of London. Series A: Mathematical and Physical Sciences, 434(1890):9–13, 1991.
  6. [9]A. N. Kolmogorov. Dissipation of energy in the locally isotropic turbulence. Proceedings of the Royal Society of London. Series A: Mathematical and Physical Sciences, 434(1890):15–17, 1991.DOI
  7. [11]A. J. Majda and A. L. Bertozzi. Vorticity and Incompressible Flow. Cambridge University Press, 2002.DOI
  8. [12]H. K. Moffatt. A brief introduction to vortex dynamics and turbulence. In Environmental Hazards: The Fluid Dynamics and Geophysics of Extreme Events, pages 1–27. World Scientific, 2011.DOI
  9. [13]A. Monin and A. Yaglom. Statistical Fluid Mechanics: Mechanics of Turbulence, volume 1 and 2. MIT Press, 1971 and 1975.
  10. [14]S. B. Pope. PDF methods for turbulent reactive flows. Progress in Energy and Combustion Science, 11(2):119–192, 1985.DOI
  11. [15]S. B. Pope. Turbulent Flows. Cambridge University Press, 2000.DOI
  12. [18]G. I. Taylor. Diffusion by continuous movements. Proceedings of the London Mathematical Society, S2-20(1):196–212, 1921.DOI
  13. [19]G. I. Taylor. Statistical theory of turbulence. Parts 1-5. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, 151(873):421–478, 1935.

Paper details

vlnc:W_arqt9rth
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents