TURBULENCE FOR THE GENERALISED BURGERS EQUATION ALEXANDRE BORITCHEV Abstract. In this survey, we review the rigorous results on turbu- lence for the generalised space-periodic Burgers equation: u +f(cid:48)(u)u = νu +η, x ∈ S1 = R/Z, t x xx studiedbyA.Biryukandtheauthorin[8,10,12,13]. Here, f issmooth and strongly convex, whereas the constant 0 < ν (cid:28) 1 corresponds to a viscosity coe(cid:30)cient. We will consider both the unforced case (η = 0) and the randomly forced case, when η is smooth in x and irregular (kick or white noise) in t. In both cases, sharp bounds for Sobolev norms of u averaged in time and in ensemble of the type Cν−δ, δ ≥ 0, with the same value of δ for upper and lower bounds, are obtained. These results yield sharp bounds for small-scale quantities characterising turbulence, con(cid:28)rming the physical predictions [7]. Abbreviations • 1d, 3d, multi-d: 1, 3, multi-dimensional • a.e.: almost everywhere • a.s.: almost surely • (GN): the Gagliardo(cid:21)Nirenberg inequality (Lemma 1.1) • i.i.d.: independent identically distributed • r.v.: random variable Introduction The generalised 1d space-periodic Burgers equation ∂u ∂u ∂2u +f(cid:48)(u) −ν = 0, ν > 0, x ∈ S1 = R/Z (1) ∂t ∂x ∂x2 (the classical Burgers equation [15] corresponds to f(u) = u2/2) is a popular model for the Navier(cid:21)Stokes equation. Indeed, both of them have similar nonlinearities and dissipative terms. Therefore, the physi- cal arguments justifying various theories of hydrodynamical turbulence Date: August 24, 2017. 1 2 ALEXANDRE BORITCHEV can usually be applied to describe the behaviour of solutions to the Burgers equation. Thus, this equation is often used as a benchmark for turbulence theories. It is also used as a benchmark for the numeri- cal methods for turbulent (cid:29)ows. For more information, see [7]. For ν (cid:28) 1 and f strongly convex, i.e. satisfying: f(cid:48)(cid:48)(x) ≥ σ > 0, x ∈ R, (2) solutions of (1) exhibit turbulent-like behaviour, called Burgers turbu- lence or (cid:16)Burgulence(cid:17) [6, 7]. To simplify the presentation, we restrict ourselves to solutions with zero mean value in space: (cid:90) u(t,x)dx = 0, ∀t ≥ 0. (3) S1 The space mean value does not change in time. Indeed, since u is 1-periodic in space, we have: (cid:90) (cid:90) (cid:90) d u(t,x)dx = − f(cid:48)(u(t,x))u (t,x)dx+ν u (t,x)dx = 0. x xx dt S1 S1 S1 If the mean value of the initial value u on S1 is equal to b (cid:54)= 0, we 0 may consider the zero mean value function v(t,x) = u(t,x+bt)−b, which is a solution of (1) with f(y) replaced by g(y) = f(y +b)−by. So the assumption (3) does not lead to a loss of generality. In this survey, we consider both the unforced equation (1) and the generalised Burgers equation with an additive forcing term, smooth in space and irregular in time (see Subsection 1.2). We summarise the estimates obtained by Biryuk and the author [8, 10, 12, 13] for the Sobolev norms as well as for the dissipation length scale and the small- scale quantities relevant for the theory of hydrodynamical turbulence: the structure functions and the energy spectrum. This survey is par- tiallybasedonthePh.D.thesisoftheauthor[11], wheresometechnical points are covered in more detail. The major di(cid:27)erence between the unforced and the white-forced gen- eralised Burgers equation is the energy picture. In the (cid:28)rst case, we have a dissipative system: the L norm is decreasing in time. Conse- 2 (cid:82) quently, the regime where the energy u2/2 dissipates fast enough S1 (which yields a time-averaged lower bound on the Sobolev norms) is transient and depends on the initial condition through a certain quan- tity D (see (15) for its de(cid:28)nition). On the contrary, in the second case, after a time needed either to dissipate energy if u is large or to supply 0 energy if u is small, we are in a quasi-stationary regime, in the sense 0 that in average on large enough time intervals, we have an approximate 3 balance between the dissipation rate −νE(cid:107)u(cid:107)2 and the constant energy 1 supply rate I . 0 For the unforced Burgers equation, some upper estimates for small- scale quantities are well-known. For example, Lemma 4.1 of our work is an analogue in the periodic setting of the one-sided Lipschitz esti- mate due to Oleinik, and the upper estimate for the structure function S ((cid:96)) follows from an estimate for the solution in the class of bounded 1 variation functions BV. For references on these classical aspects of the theory of scalar conservation laws, see [20, 45, 49]. For some upper es- timates for small-scale quantities, see [38, 53]. To our best knowledge, rigorous lower estimates were not known before Biryuk’s and our work. Theresearchonthesmall-scalebehaviourofsolutionsforthe(forced) generalisedBurgersequationismotivatedbytheproblemofturbulence. It has been inspired by the pioneering works of Kuksin, who obtained lower and upper estimates for Sobolev norms by negative powers of the viscosity for a large class of equations (see [41, 42] and references in [42]). For more recent results obtained by Kuksin, Shirikyan and oth- ers for the 2D Navier(cid:21)Stokes equation, see the book [43] and references therein. The estimates for Sobolev norms and for small-scale quantities pre- sented in our work are asymptotically sharp in the sense that viscosity enters lower and upper bounds at the same negative power. Such esti- mates are not available for the more complicated equations considered in [41, 42, 43]. This survey is also concerned with the problem of the invariant mea- sure for the stochastic generalised Burgers equation. This problem has been treated prevously for the nonlinearity uu by Sinai [51, 52] and in x the inviscid limit ν → 0 by E, Khanin, Mazel and Sinai [22]; see also [28, 31]. Here we present a simple approach to this problem, described in [13], which consists in using L -contractivity of the (cid:29)ow correspond- 1 ing to the equation, and a coupling argument. We do not consider other aspects of Burgulence, such as the inviscid limit, the behaviour of solutions for spatially rough forcing and the noncompact setting. We refer the reader to the survey by Bec and Khanin [7], which is concerned with physical aspects of the theory of Burgulence, and to the survey by Bakhtin [3], which discusses related probabilistic and ergodic results. Organisation of the paper: We begin by introducing the notation and setup in Section 1. In Section 2, we present the K41 theory as well as the physical predictions for Burgulence. In Section 3, we formulate the main results. 4 ALEXANDRE BORITCHEV InSection4, weconsiderthesolutionu(t,x)oftheunforcedequation (1). In Subsection 4.1, we begin by recalling the upper estimate for the quantity max u (s,x), t ≥ 1. x s∈[t,t+1], x∈S1 Using this bound, we get upper and lower estimates for the Sobolev norms of u. In Subsection 4.2 we study the implications of our re- sults for Burgulence theory. Namely, we give sharp upper and lower bounds for the dissipation length scale, the increments and the spec- tral asymptotics for the (cid:29)ow u(t,x), which hold uniformly for ν ≤ ν . 0 The quantity ν > 0 depends only on f and on the initial condition. 0 These results justify rigorously the physical predictions for small-scale quantities which characterise Burgulence. Inthetwolastsectionsofthepaper, weconsidertherandomlyforced generalised Burgers equation. In Section 5, we obtain analogues of the results in Section 4, which also con(cid:28)rm the corresponding physical pre- dictions [7]. Section 6 is concerned with the stationary measure. 1. Notation and setup All functions which we consider in this paper are real-valued, except in Section 2, where vectors in R3 are written in bold script. 1.1. Functional spaces and Sobolev norms. Consider a zero mean valueintegrablefunctionv onS1. Forp ∈ [1,∞], wedenoteitsL norm p by|v| . TheL normisdenotedby|v|, and(cid:104)·,·(cid:105)standsfortheL scalar p 2 2 product. From now on L , p ∈ [1,∞], denotes the space of zero mean p value functions in L (S1). Similarly, C∞ is the space of C∞-smooth p zero mean value functions on S1. For a nonnegative integer m and p ∈ [1,∞], Wm,p stands for the Sobolev space of zero mean value functions v on S1 with (cid:28)nite homo- geneous norm (cid:12) (cid:12) (cid:12)dmv(cid:12) |v| = (cid:12) (cid:12) . m,p (cid:12)dxm(cid:12) p In particular, W0,p = L for p ∈ [1,∞]. For p = 2, we denote Wm,2 by p Hm and abbreviate the corresponding norm as (cid:107)v(cid:107) . m Since the length of S1 is 1, we have: |v| ≤ |v| ≤ |v| ≤ |v| ≤ ··· ≤ |v| ≤ |v| ≤ ... 1 ∞ 1,1 1,∞ m,1 m,∞ We recall a version of the classical Gagliardo(cid:21)Nirenberg inequality (see [21, Appendix]): 5 Lemma 1.1. For a smooth zero mean value function v on S1, |v| ≤ C|v|θ |v|1−θ, β,r m,p q where m > β ≥ 0, and r is de(cid:28)ned by 1 (cid:16) 1(cid:17) 1 = β −θ m− +(1−θ) , r p q under the assumption θ = β/m if p = 1 or p = ∞, and β/m ≤ θ < 1 otherwise. The constant C depends on m,p,q,β,θ. From now on, we will refer to this inequality as (GN). For any s ≥ 0, Hs stands for the Sobolev space of zero mean value functions v on S1 with (cid:28)nite norm (cid:16)(cid:88) (cid:17)1/2 (cid:107)v(cid:107) = (2π)s |k|2s|vˆ(k)|2 , (4) s k∈Z where vˆ(k) are the complex Fourier coe(cid:30)cients of v(x). For an integer s = m, this norm coincides with the previously de(cid:28)ned Hm norm. For s ∈ (0,1), (cid:107)v(cid:107) is equivalent to the norm s (cid:32) (cid:33)1/2 (cid:90) (cid:16)(cid:90) 1 |v(x+(cid:96))−v(x)|2 (cid:17) (cid:48) (cid:107)v(cid:107) = d(cid:96) dx (5) s (cid:96)2s+1 S1 0 (see [1, 54]). Subindices t and x, which can be repeated, denote partial di(cid:27)erenti- ation with respect to the corresponding variables. We denote by v(m) the m-th derivative of v in the variable x. For brevity, the function v(t,·) is denoted by v(t). 1.2. Well-posedness and di(cid:27)erent types of forcing. In Section 4, we consider the unforced equation (1) with a C∞-smooth initial con- dition u . This equation has a unique solution in C∞: see for instance 0 [39, Chapter 5]. In Section 5, we consider the generalised Burgers equation with two di(cid:27)erent types of additive forcing in the right-hand side, taking again a C∞-smooth initial condition u . Since the forcing always has zero 0 mean value in space and the initial condition satis(cid:28)es (3), its solutions satisfy (3) for all time. First, we consider the kick force. We begin by providing the space L 2 with the Borel σ-algebra (Ω,F). Then we consider an L -valued r.v. 2 ζ = ζω on a probability space (Ω,F,P). We suppose that ζ satis(cid:28)es the following three properties. 6 ALEXANDRE BORITCHEV (i) (Non-triviality) P(ζ ≡ 0) < 1. (ii) (Finiteness of moments for Sobolev norms) For every m ≥ 0, we have: I = E(cid:107)ζ(cid:107)2 < +∞. m m (iii) (Vanishing of the expected value) Eζ ≡ 0. It is not di(cid:30)cult to construct explicitly ζ satisfying (i)-(iii). One possibilty is to suppose that the real Fourier coe(cid:30)cients of ζ, de(cid:28)ned for k > 0 by √ (cid:90) √ (cid:90) a (ζ) = 2 cos(2πkx)u(x); b (ζ) = 2 sin(2πkx)u(x), (6) k k S1 S1 are independent r.v. with zero mean value and exponential moments tending to 1 fast enough as k → +∞. Now let ζ = ζω, i ∈ N be i.i.d. r.v.’s having the same distribution as i i ζ. The sequence (ζ ) is a r.v. de(cid:28)ned on a probability space which i i≥1 is a countable direct product of copies of Ω. From now on, this space will itself be called Ω. The meaning of F and P changes accordingly. For ω ∈ Ω, the kick force ξω is by de(cid:28)nition the distribution de(cid:28)ned by +∞ (cid:88) ξω(t,x) = δ ζω(x), t=i i i=1 where δ denotes the Dirac measure at the time moment i. t=i The kick-forced equation corresponds to the case where, in the right- hand side of (1), 0 is replaced by the kick force: ∂u ∂u ∂2u +f(cid:48)(u) −ν = ξω. (7) ∂t ∂x ∂x2 This means that for integers i ≥ 1, at the moments i the solution u(x) instantly increases by the kick ζω(x), and that between these moments i u solves (1). We make the additional assumption that the solution is a right-continuous function of time. Existence and uniqueness of spatially smooth solutions to (7) follows directly from the corresponding fact for the unforced equation. The other type of forcing considered here is the white force. Heuris- tically this force corresponds to a scaled limit of kick forces with more 7 and more frequent kicks. We provide each space Wm,p with the Borel σ-algebra. Then we consider an L -valued Wiener process 2 w(t) = wω(t), ω ∈ Ω, t ≥ 0, de(cid:28)ned on a complete probability space (Ω, F, P), and an adapted (cid:28)ltration {F , t ≥ 0} (i.e., for t ≥ 0, w(t) is F -measurable, and F t t t and the σ-algebra generated by the r.v.’s w(t + s) − w(t), s ≥ 0 are independent). We assume that for each m and each t ≥ 0, w(t) ∈ Hm, almost surely. That is, for ζ,χ ∈ L , 2 E((cid:104)w(s),ζ(cid:105)(cid:104)w(t),χ(cid:105)) = min(s,t)(cid:104)Qζ,χ(cid:105), where Q is a symmetric operator which de(cid:28)nes a continuous mapping Q : L → Hm for every m. Thus, w(t) ∈ C∞ for every t, almost surely. 2 From now on, we rede(cid:28)ne the Wiener process so that this property holds for all ω ∈ Ω. We will denote w(t)(x) by w(t,x). For m ≥ 0, we denote by I the quantity m I = Tr (Q) = E(cid:107)w(1)(cid:107)2 . m Hm m FormoredetailsonWienerprocessesinHilbertspaces, see[18, Chapter 4] and [44]. It is not di(cid:30)cult to construct w(t) explicitly. For instance, we could consider the particular case of a (cid:16)diagonal(cid:17) noise: √ √ (cid:88) (cid:88) w(t) = 2 a w (t)cos(2πkx)+ 2 b w˜ (t)sin(2πkx), k k k k k≥1 k≥1 where w (t), w˜ (t), k > 0, are standard independent Wiener processes k k and (cid:88) I = (a2 +b2)(2πk)2m < ∞ m k k k≥1 for each m. From now on, dw(s) denotes the stochastic di(cid:27)erential corresponding to the Wiener process w(s) in the space L . 2 Now (cid:28)x m ≥ 0. By Fernique’s Theorem [44, Theorem 3.3.1], there exist λ ,C > 0 such that m m (cid:16) (cid:17) Eexp λ (cid:107)w(T)(cid:107)2 /T ≤ C , T ≥ 0. (8) m m m Therefore by Doob’s maximal inequality for in(cid:28)nite-dimensional sub- martingales [18, Theorem 3.8. (ii)] we have: (cid:16) p (cid:17)p E sup (cid:107)w(t)(cid:107)p ≤ E(cid:107)w(T)(cid:107)p < +∞, (9) m p−1 m t∈[0,T] 8 ALEXANDRE BORITCHEV for any T > 0 and p ∈ (1,∞). The white-forced equation is obtained by replacing 0 by the weak derivativeηω = ∂wω/∂tintheright-handsideof(1). Here,wω(t), t ≥ 0 is the Wiener process de(cid:28)ned above. Definition 1.2. For T ≥ 0, we say that an H1-valued process u(t,x) = uω(t,x) is a solution of the equation ∂uω ∂uω ∂2uω +f(cid:48)(uω) −ν = ηω (10) ∂t ∂x ∂x2 for t ≥ T if: (i) For every t ≥ T, ω (cid:55)→ uω(t,·) is F -measurable. t (ii) For every ω, t (cid:55)→ uω(t,·) is continuous in H1 for t ≥ T and satis(cid:28)es (cid:90) t(cid:16) 1 (cid:17) uω(t) =uω(T)− νLuω(s)+ B(uω)(s) ds 2 T +wω(t)−wω(T), (11) where B(u) = 2f(cid:48)(u)u ; L = −∂ . x xx For brevity, solutions for t ≥ 0 will be referred to as solutions. Existence and uniqueness of smooth solutions to (10) is proved by the mild solution technique (cf. [19, Chapter 14]). Since the forcing and the initial condition are smooth in space, the mapping t (cid:55)→ u(t) is time-continuous in Hm for every m, and t (cid:55)→ u(t)−w(t) has a space derivative in C∞ for all t, a.s. Nowconsider,forasolutionu(t,x)of(10),thefunctionalG (u(t)) = m (cid:107)u(t)(cid:107)2 and apply It(cid:244)’s formula [18, Theorem 4.17]: m (cid:90) t (cid:107)u(t)(cid:107)2 =(cid:107)u (cid:107)2 − (cid:0)2ν(cid:107)u(s)(cid:107)2 +(cid:104)Lmu(s), B(u)(s)(cid:105)(cid:1)ds+tI m 0 m m+1 m 0 (cid:90) t +2 (cid:104)Lmu(s), dw(s)(cid:105) 0 (we recall that I = Tr(Q ).) Consequently, m m d E(cid:107)u(t)(cid:107)2 = −2νE(cid:107)u(t)(cid:107)2 −E (cid:104)Lmu(t), B(u)(t)(cid:105)+I . dt m m+1 m As (cid:104)u, B(u)(cid:105) = 0, for m = 0 this relation becomes d E|u(t)|2 = I −2νE(cid:107)u(t)(cid:107)2. (12) dt 0 1 9 1.3. Notation and agreements. When considering a Sobolev norm in Wm,p, the quantity γ = γ(m,p) denotes max(0, m−1/p). In Subsection 2.1, v(t,x) denotes the velocity of a 3d (cid:29)ow with pe- riod 1 in each spatial coordinate. In the whole paper, u(t,x) denotes a solution of the generalised Burgers equation with a given initial condi- tion u = u(0,·). In Section 4, we deal with the equation (1) under the 0 assumptions (2-3). In Section 5 we deal with the equation (10) under the assumptions (2-3) and under the additional condition ∀m ≥ 0, ∃h ≥ 0, C > 0 : |f(m)(x)| ≤ C (1+|x|)h, x ∈ R, (13) m m where h = h(m) is a function such that 1 ≤ h(1) < 2 (14) (the lower bound on h(1) follows from (2)). The results in that section also hold for the kicked equation (7), under the same assumptions as for (10), except (14), which is unnecessary. WhenweconsidertherandomlyforcedgeneralisedBurgersequation, PetEdenote,respectively,theprobabilityandtheexpectedvaluewith respect to the probability measure Ω (cf. Section 1.2). All quantities denoted by C with sub- or superindices are positive and nonrandom. Unless otherwise stated, they depend only on the following parameters: • When dealing with the K41 theory, the statistical properties of the forcing. • When studying the unforced generalised Burgers equation, the function f determining the nonlinearity f(cid:48)(u)u , as well as the x parameter D = max(|u |−1, |u | ) (15) 0 1 0 1,∞ which characterises how generic the initial condition is. • When studying the randomly forced generalised Burgers equa- tion, the function f determining the nonlinearity f(cid:48)(u)u , as x well as the statistical properties of the forcing. In the case of a kick force, by statistical properties we mean the distribution function of the i.i.d. r.v.’s ζ . In the case of a white force, i we mean the correlation operator Q for the Wiener process w de(cid:28)ning the random forcing. In particular, these quantities never depend on the viscosity coe(cid:30)cient ν. Constants which also depend on parameters a ,...,a are denoted 1 k 10 ALEXANDRE BORITCHEV a1,...,ak by C(a ,...,a ). By X (cid:46) Y we mean that X ≤ C(a ,...,a )Y. 1 k 1 k The notation X a1,∼...,ak Y stands for a1,...,ak a1,...,ak Y (cid:46) X (cid:46) Y. In particular, X (cid:46) Y and X ∼ Y mean that X ≤ CY and C−1Y ≤ X ≤ CY, respectively. Note that this notation is never used wit the parameter ν: in other words, dependence on the viscosity is always explicitly speci(cid:28)ed. We use the notation g− = max(−g,0) and g+ = max(g,0). InSubsection2.1, thebrackets(cid:104)·(cid:105)denotetheexpectedvalue. Forthe meaningofthebrackets{·}, seeSubsection4.1inthedeterministiccase (where they correspond to averaging in time) and Subsection 5.3 in the random case (where they correspond to averaging in time and taking the expected value). The de(cid:28)nitions of the relevant ranges and the length scales, as well as of the small-scale quantities, i.e. the structure functions S and S = S and the spectrum E(k) depend on the p,α p,1 p setting: see Subsections 2.1, 2.2, 4.2 and 5.3. 2. Turbulence and the Burgers equation 2.1. Turbulence, K41theory, intermittency. Itiswell-knownthat giving a precise de(cid:28)nition of turbulence is problematic. However, some features are generally recognised as characteristic of turbulence: many degrees of freedom, unpredictability/chaos, (small-scale) irregularity... For a more detailed discussion, see [27, 55]. Here, we will only present (in a slightly modi(cid:28)ed form) the vocabulary of the theory of turbulence which is relevant to the study of the Burgers model. In particular, we will proceed as if the (cid:29)ow v(t,x) under consideration is periodic in space, without concerning ourselves with the physical relevance of K41 in this setting. Without loss of generality, we may assume that v is 1- periodic in each coordinate x ,x ,x . Let us denote by ν the viscosity 1 2 3 coe(cid:30)cient; we only consider the turbulent regime 0 < ν (cid:28) 1. We de(cid:28)ne the space scale as the inverse of the frequency under con- sideration. In particular, the Fourier coe(cid:30)cients vˆ(k) for large values of k or, in the physical space, the increments v(x+r)−v(x) for small values of r, are prototypical small-scale quantities. The theory which may be considered as a starting point for the modern study of turbulence is essentially contained in three articles by Kolmogorov which have been published in 1941 [33, 34, 35]. Thus, it is referred to as the K41 theory. The philosophy behind K41 is that although large-scale characteris- tics of a turbulent (cid:29)ow are clearly individual (depending on the forcing
Description: