betaFeedback

WorksPhysics and Astronomy

A Set-theoretic Approach to Regularity of 3D Decaying Turbulence

Abstract

This article first proves a differential version of the energy equality for the 3D decaying turbulence. It is weaker than the integral equality, as stronger conditions of continuity are needed when integrating the differential equality. Using the differential energy equality, a simple criterion for a weak solution to satisfy the strong energy inequality is proposed. Namely, the monotone-decreasing kinetic energy in time is equivalent to the LH condition. Lastly, motivated by the preceding criterion, we propose an original perspective in understanding the existing concepts of turbulence, which uses set-theoretic tools such as partial ordering and Zorn's Lemma. The partial ordering in time defines a set-theoretic characterization of regularity for weak solutions.

2020 MSC : 76-02, 76D03, 76D05

We consider the Navier-Stokes equations for incompressible viscous flow on the 3D torus, in the absence of an external force:

∂u∂t+(u⋅∇)u−Δu=−∇p(1)\frac{\partial u}{\partial t} + (u \cdot\nabla)u - \Delta u = -\nabla p \tag*{(1)}
∇⋅u=0(2)\nabla\cdot u = 0 \tag*{(2)}

where u(x,t)u(x,t) is the velocity vector, pp is the pressure, and we have chosen units in which the density and kinematic viscosity are equal to one, and the torus T3=(R/(2πZ))3\mathbb{T}^{3} = \left(\mathbb{R}/(2\pi\mathbb{Z})\right)^{3}. Regularity of weak solutions of Eq. (1) is a central topic in mathematical fluid mechanics. An important result is that a weak solution satisfying the strong energy inequality is smooth possibly except for a set of singular points in time contained in a bounded interval.

This article proposes a simple criterion for a weak solution to satisfy the strong energy inequality, namely monotone-decreasing kinetic energy in time. Motivated by this result, we propose a new perspective in understanding regularity of turbulence, which uses set-theoretic tools.

Such findings may be seen as analysis of turbulence in the non-equilibrium state. Weak solutions with the strong energy inequality are known to be smooth for all sufficiently large time. Thus, it can be said that they converge to an equilibrium in the long run. However, the behavior of turbulence in finite time still remains as an open problem. Our set-theoretic approach provides a novel way to understand potential singularities in finite time.

Mathematical Framework and Definitions

In this section, we briefly present basic concepts, with [3] as the main reference.

Function Spaces

Definition 1. Let H\mathcal{H} and V\mathcal{V} be the completion of divergence-free smooth vector fields on T3\mathbb{T}^{3} in [L2(T3)]3\left[L^{2}(\mathbb{T}^{3})\right]^{3} and [H1(T3)]3\left[H^{1}(\mathbb{T}^{3})\right]^{3} respectively. By V′\mathcal{V}', we denote the dual space of V\mathcal{V}.

Definition 2. Let I⊂RI \subset\mathbb{R} be a bounded interval, and XX be a separable Banach space. For p∈[1,∞]p \in[1,\infty], we define the Bochner space

Lp(I,X)(3)L^{p}(I,X) \tag*{(3)}

to be the LpL^{p} space of functions u(t)u(t) taking values in XX. If u~:I→X\tilde{u}: I \to X satisfies u~=u\tilde{u}=u almost everywhere on II, it is called a version of uu.

Weak time-derivatives for Bochner functions is defined in [3 Definition 1.30] and some of their basic properties are discussed in [3 Lemma 1.31]. In particular, if u∈L1(0,T;X)u \in L^{1}(0,T;X) has a weak derivative ∂tu\partial_{t}u, there exists ξ∈X\xi\in X such that

u~(t)=ξ+∫0t∂tu(s) ds,(4)\tilde{u}(t)=\xi+\int_{0}^{t}\partial_{t}u(s)\,\mathrm{d}s, \tag*{(4)}

is a version of uu.

In this case, it turns out that ∂tu\partial_{t}u is a Fréchet derivative of u~\tilde{u} in the following sense:

∥u~(t+α)−u~(t)−α ∂tu(t)∥X=o(α) as α→0(5)\left\|\tilde{u}(t+\alpha)-\tilde{u}(t)-\alpha\,\partial_{t}u(t)\right\|_{X}=o(\alpha)\ \text{as }\alpha\to0 \tag*{(5)}

for almost every t∈[0,T]t \in[0,T]. Indeed,

u~(t+α)−u~(t)−α∂tu(t)=∫tt+α(∂tu(s)−∂tu(t)) ds\widetilde{u}(t+\alpha)-\widetilde{u}(t)-\alpha\partial_tu(t)=\int_t^{t+\alpha}\left(\partial_tu(s)-\partial_tu(t)\right)\,\mathrm{d}s

and it is known that

lim⁡α→01α∫tt+α∥∂tu(s)−∂tu(t)∥X ds=0\lim_{\alpha\to0}\frac{1}{\alpha}\int_t^{t+\alpha}\left\|\partial_tu(s)-\partial_tu(t)\right\|_X\,\mathrm{d}s=0

for almost every tt. (This is a generalized form of the Lebesgue differentiation theorem. See, for example, [2 Thm 3.4].)

Weak solutions and Their Properties

A precise definition of a weak solution of the Navier-Stokes equation Eq. (1) is presented in [3 Chapter 3]. Here we only collect the properties necessary for this paper.

Let uu be a weak solution, then uu satisfies the following:

  • u∈L2(0,T;V)∩L∞(0,T;H)u \in L^2(0,T;V) \cap L^\infty(0,T;H)

  • ∂tu\partial_tu, (u⋅∇)u∈L4/3(0,T;V′)(u \cdot\nabla)u \in L^{4/3}(0,T;V')

  • ∂tu−Δu+(u⋅∇)u=0\partial_tu-\Delta u+(u \cdot\nabla)u=0 as elements of L4/3(0,T;V′)L^{4/3}(0,T;V')

for all T∈(0,∞)T \in(0,\infty). The first and third properties are from [3 Definition 3.3], while proof of the second one can be found in [3 Lemma 3.4 and Lemma 3.7]. Via Eq. (4), it is assumed uu is Fréchet-differentiable almost everywhere when regarded as a V′V'-valued mapping.

Next, we introduce a sub-class of weak solutions:

Definition 3. A weak solution is said to satisfy the strong energy inequality if

12∥u(t)∥2+∫st∥∇u∥2≤12∥u(s)∥2(∀t>s),(6)\frac{1}{2}\lVert u(t)\rVert^2+\int_s^t\lVert\nabla u\rVert^2\leq\frac{1}{2}\lVert u(s)\rVert^2 \qquad(\forall t>s), \tag*{(6)}

for s=0s=0 and almost all times s∈(0,∞)s\in(0,\infty). Such weak solutions are sometimes called Leray-Hopf.

It is well-known that every Leray-Hopf weak solution is almost smooth and in particular, smooth for all large time tt. For more details, one may consult original papers, cf. [1], or modern textbooks such as [3 Theorem 8.14] and references therein.

Main Results

In this section, the main results are stated. We start with the following lemma, namely the differential energy equality:

Lemma 4. Let uu be a weak solution to Eq. (1). Then

⟨∂tu,u⟩=−∥∇u∥2(7)\langle\partial_t u,u\rangle=-\lVert\nabla u\rVert^2 \tag*{(7)}

for almost all tt, where the bracket denotes dual pairing between V′V' and VV.

Proof. According to § 1.2,

∂tu−Δu+(u⋅∇)u=0 as elements of V′(8)\partial_t u-\Delta u+(u\cdot\nabla)u=0\text{ as elements of }V' \tag*{(8)}

holds for almost all tt while u∈Vu\in V for almost tt. Therefore, we evaluate Eq. (8) against uu for almost all tt to obtain

⟨∂tu,u⟩=⟨Δu,u⟩−⟨(u⋅∇)u,u⟩(9)\langle\partial_tu,u\rangle=\langle\Delta u,u\rangle-\langle(u\cdot\nabla)u,u\rangle \tag*{(9)}

where the first term in the rhs of Eq. (9) is equal to −∥∇u∥2-\lVert\nabla u\rVert^2 by construction while the second term in the rhs vanishes by [3 Lemma 3.2].

Note that the version of a weak solution uu (with initial data u0∈Hu_0\in\mathcal{H}) as in Eq. (4) satisfies u(t)→u(0)u(t)\to u(0) in V′V' as t→0+t\to0^+. By [3 Theorem 3.8] together with the fact that u0∈H⊂V′u_0\in\mathcal{H}\subset V', we observe that u(0)u(0) in fact belongs to H\mathcal{H} and coincides with u0u_0.

Theorem 5. Let uu be a weak solution of Eq. (1) with initial data u0∈Hu_0\in\mathcal{H}. Then, the following are equivalent:

  1. There exists a version uˉ\bar{u} of uu such that t↦∥uˉ(t)∥2t\mapsto\lVert\bar{u}(t)\rVert^2 is monotone-decreasing and ∥uˉ(0)∥=∥u0∥\lVert\bar{u}(0)\rVert=\lVert u_0\rVert.

  2. uu is Leray-Hopf.

Proof. First, we show that (1) implies (2).

Let uˉ\bar{u} be a version of uu such that t↦∥uˉ(t)∥2t\mapsto\lVert\bar{u}(t)\rVert^2 is monotone-decreasing with ∥uˉ(0)∥=∥u0∥\lVert\bar{u}(0)\rVert=\lVert u_0\rVert. Then the following conditions hold almost everywhere:

(i) uˉ(t)=u(t)\bar{u}(t)=u(t) as elements of VV. This holds for a.e. tt because uˉ\bar{u} is a version of uu, and because u∈L2Vu\in L^2V.

(ii) t↦∥uˉ(t)∥2t\mapsto\lVert\bar{u}(t)\rVert^2 is differentiable in the classical sense. This holds a.e. due to monotonicity [4 Theorem 5.2].

(iii) u(t)u(t) is Fréchet differentiable. See Eq. (5).

(iv) ⟨∂tu(t),u(t)⟩=−∥∇u(t)∥2\langle\partial_tu(t),u(t)\rangle=-\lVert\nabla u(t)\rVert^2 (Lemma 4).

Let RR be the (dense) subset of (0,∞)(0,\infty) on which all of these conditions hold. Taking tt and t+αt+\alpha in RR, a direct computation leads to

∥u(t+α)∥2−∥u(t)∥2=∥u(t+α)−u(t)∥2+2⟨u(t+α)−u(t)−α∂tu(t),u(t)⟩+2α⟨∂tu(t),u(t)⟩(10)\begin{aligned} \lVert u(t+\alpha)\rVert^2-\lVert u(t)\rVert^2 ={}&\lVert u(t+\alpha)-u(t)\rVert^2\\ &+2\langle u(t+\alpha)-u(t)-\alpha\partial_tu(t),u(t)\rangle +2\alpha\langle\partial_tu(t),u(t)\rangle \tag*{(10)} \end{aligned}

where the brackets are dual pairing between V′V' and VV. Note that

∣⟨u(t+α)−u(t)−α ∂tu(t),u(t)⟩∣≤∥u(t+α)−u(t)−α ∂tu(t)∥V′⋅∥u(t)∥V(11)\begin{aligned} \left|\langle u(t+\alpha)-u(t)-\alpha\,\partial_tu(t),u(t)\rangle\right| &\leq\|u(t+\alpha)-u(t)-\alpha\,\partial_tu(t)\|_{V'}\cdot\|u(t)\|_V \tag*{(11)} \end{aligned}

and the bound is o(∣α∣)o(|\alpha|) almost everywhere due to Fréchet differentiability as in Eq. (5).

Let t0∈Rt_0\in\mathcal{R}. Since we know that the derivative of ∥uˉ(t)∥2\|\bar{u}(t)\|^2 at t0t_0 exists, we can compute it by taking α→0\alpha\to0 through values with t0+α∈Rt_0+\alpha\in\mathcal{R}. Assume first that α>0\alpha>0 and divide Eq. (10) by α\alpha, recalling that ∥uˉ(t)∥2=∥u(t)∥2\|\bar{u}(t)\|^2=\|u(t)\|^2 for t∈Rt\in\mathcal{R}. The first term on the rhs of Eq. (10) is nonnegative, the second term goes to zero because of Eq. (11), and the third term is 2⟨∂tu,u⟩2\langle\partial_tu,u\rangle. It follows that

ddt∥uˉ(t)∥2∣t=t0≥2⟨∂tu(t0),u(t0)⟩.\left.\frac{d}{dt}\|\bar{u}(t)\|^2\right|_{t=t_0}\geq2\langle\partial_tu(t_0),u(t_0)\rangle.

Repeating the same procedure with α\alpha negative yields the opposite inequality. (This proves, incidentally, that the derivative of the first term on the rhs of Eq. (10) is zero almost everywhere.) Using (i)(i) and (iv)(iv), it follows that

ddt∥uˉ(t)∥2=−2∥∇uˉ(t)∥2\frac{d}{dt}\|\bar{u}(t)\|^2=-2\|\nabla\bar{u}(t)\|^2

almost everywhere. Integrating, we infer that

∥uˉ(t)∥2−∥uˉ(s)∥2≤−2∫st∥∇u(r)∥2 dr(12)\|\bar{u}(t)\|^2-\|\bar{u}(s)\|^2\leq-2\int_s^t\|\nabla u(r)\|^2\,dr \tag*{(12)}

for all 0≤s<t<∞0\leq s<t<\infty [4 Theorem 5.2]. Since ∥uˉ(t)∥=∥u(t)∥\|\bar{u}(t)\|=\|u(t)\| almost everywhere, including t=0t=0, we deduce from Eq. (12) that

∥u(t)∥2−∥u(s)∥2≤−2∫st∥∇u(r)∥2 dr(13)\|u(t)\|^2-\|u(s)\|^2\leq-2\int_s^t\|\nabla u(r)\|^2\,dr \tag*{(13)}

for almost all t>st>s, for almost all ss, including s=0s=0. It follows from [3 p.99-100] that we can extend Eq. (13) to hold for all t>st>s, for almost all ss, including s=0s=0. This completes the proof that (1)⇒(2)(1)\Rightarrow(2).

We now show that (2)⇒(1)(2)\Rightarrow(1). Let

f(t)=−∥u(t)∥2.f(t)=-\|u(t)\|^2.

Then the strong energy inequality implies that for almost all s∈[0,∞)s\in[0,\infty), including s=0s=0,

f(t)≥f(s)(∀t>s).(14)f(t)\geq f(s)\qquad(\forall t>s). \tag*{(14)}

To prove (2)⇒(1)(2)\Rightarrow(1), we need only show that any such ff possesses a version fˉ\bar{f} such that Eq. (14) holds for all s≥0s\geq0 and fˉ(0)=f(0)\bar{f}(0)=f(0).

Let SS be the set for which Eq. (14) holds. Define

fˉ(s)=inf⁡s′≥s,s′∈Sf(s′).\bar{f}(s)=\inf_{\substack{s'\geq s,\,s'\in S}} f(s').

where the infimum exists because ff is monotonic on SS so that f(s)≥f(0)f(s) \ge f(0) for all s∈Ss \in S. Note also that fˉ(s)=f(s)\bar{f}(s) = f(s) trivially when s∈Ss \in S. To show that Eq. (14) holds for all ss for fˉ\bar{f}, choose s0∈(s,t)∩Ss_0 \in(s,t) \cap S. Then

fˉ(s)=inf⁡s′≥ss′∈Sf(s′)≤f(s0)≤inf⁡s′′≥ts′′∈Sf(s′′)=fˉ(t).\bar{f}(s)=\inf_{\substack{s'\ge s\\s'\in S}}f(s')\le f(s_0)\le\inf_{\substack{s''\ge t\\s''\in S}}f(s'')=\bar{f}(t).

This concludes the proof of the first theorem.

Next, the notion of regular and singular times is introduced:

Definition 6. Let uu be a weak solution. t0∈(0,∞)t_0 \in(0,\infty) is called a regular time if uu is smooth in some neighborhood of t0t_0 in (0,∞)(0,\infty). Otherwise, t0t_0 is called a singular time.

This is just [3], Definition 8.2 except that uu is not required to be Leray-Hopf and the initial time (t0=0t_0=0) is not considered. Many of fundamental results on regularity of weak solutions, including [3], Theorem 8.14 and Theorem 8.17, assume the strong energy inequality as an input. On the other hand, this paper aims to analyze cases where this inequality fails to hold. The initial time is excluded to avoid technical complications of dealing with smoothness at an endpoint of an interval.

Definition 6, together with Theorem 5, motivates the following binary relation:

Definition 7. Let uu be a weak solution and I⊂(0,∞)I \subset(0,\infty) be an open interval. For s,t∈Is,t \in I, we write s⪯ts \preceq t if

  1. s=ts=t or

  2. s<ts<t and there exist bounded open intervals (s1,s2)(s_1,s_2) and (t1,t2)(t_1,t_2) in II such that

    • s2<t1s_2<t_1

    • s∈(s1,s2)s\in(s_1,s_2) and t∈(t1,t2)t\in(t_1,t_2)

    • ∥u(t′)∥≤∥u(s′)∥\lVert u(t')\rVert\le\lVert u(s')\rVert for almost all s′∈(s1,s2)s'\in(s_1,s_2), for almost all t′∈(t1,t2)t'\in(t_1,t_2).

Note that this definition is independent of any version of uu.

Recall that a binary relation is a partial ordering if it is reflexive, antisymmetric and transitive.

Proposition 8. The relation ⪯\preceq is a partial ordering on II.

Proof. Reflexivity and antisymmetry are obvious from the definition. For transitivity, let s⪯ts\preceq t and t⪯wt\preceq w. We only need to consider the case s<t<ws<t<w.

Let (s1,s2)(s_1,s_2), (t1,t2)(t_1,t_2) and (w1,w2)(w_1,w_2) be intervals as in Definition 7. Denote by W⊂(w1,w2)W\subset(w_1,w_2) the set of w′∈(w1,w2)w'\in(w_1,w_2) satisfying ∥u(w′)∥≤∥u(t′)∥\lVert u(w')\rVert\le\lVert u(t')\rVert for almost all t′∈(t1,t2)t'\in(t_1,t_2). For each w′∈Ww'\in W, let Tw′⊂(t1,t2)T_{w'}\subset(t_1,t_2) be the set of all such t′t'.

Similarly, we introduce the notations T⊂(t1,t2)T \subset(t_1,t_2) and St′⊂(s1,s2)S_{t'} \subset(s_1,s_2) for each t′∈Tt' \in T.

Consider

⋃t′∈Tw′∩TSt′\bigcup_{t' \in T_{w'} \cap T} S_{t'}

for each w′∈Ww' \in W. Since (t1,t2)−(Tw′∩T)(t_1,t_2) - (T_{w'} \cap T) is a null set, Tw′∩TT_{w'} \cap T is nonempty. Any s′∈⋃t′∈Tw′∩TSt′s' \in\bigcup_{t' \in T_{w'} \cap T} S_{t'} belongs to St′S_{t'} for some t′∈Tw′∩Tt' \in T_{w'} \cap T so that ∥u(t′)∥≤∥u(s′)∥\lVert u(t') \rVert\le\lVert u(s') \rVert. Moreover, ∥u(w′)∥≤∥u(t′)∥\lVert u(w') \rVert\le\lVert u(t') \rVert by definition of Tw′T_{w'}, so that we finally obtain ∥u(w′)∥≤∥u(s′)∥\lVert u(w') \rVert\le\lVert u(s') \rVert. (s1,s2)−⋃t′∈Tw′∩TSt′(s_1,s_2) - \bigcup_{t' \in T_{w'} \cap T} S_{t'} is clearly a null set.

Recall that if s∈Is \in I is a maximal element with respect to ⪯\preceq, there is no t∈It \in I satisfying both s≠ts \ne t and s⪯ts \preceq t. The following proposition shows that a maximal element with respect to ⪯\preceq is actually a singular time:

Proposition 9. Let uu be a weak solution. If t0t_0 is a maximal element in an open interval with respect to ⪯\preceq, it is a singular time.

Proof. We prove the contrapositive. Let t0t_0 be a regular time. By definition, uu is smooth in a neighborhood of t0t_0 and we may therefore find an open interval UU containing t0t_0 in which uu satisfies the strong energy inequality (in fact equality). That is, t→∥u(t)∥t \to\lVert u(t) \rVert is monotone-decreasing on UU.

Now, assume that t0t_0 is a maximal element in some open interval II with respect to ⪯\preceq. By setting UU to be sufficiently small, we may have U⊂IU \subset I. The above paragraph then implies that any t∈Ut \in U greater than t0t_0 satisfies t0⪯tt_0 \preceq t, which is contradictory to maximality of t0t_0. Therefore, t0t_0 cannot be a maximal element in any open interval.

Remark 10. Unfortunately, it has not been possible to prove the converse. In fact, if the converse of Proposition 9 holds, the kinetic energy of a non-smooth weak solution cannot be monotone-decreasing. Therefore, Theorem 5 implies that every Leray-Hopf solution must be smooth on whole of (0,∞)(0,\infty).

In other words, proof of the converse of Proposition 9 is essentially to solve the Millenium Prize Problem.

The following proposition has been motivated by Zorn’s Lemma:

Proposition 11. Let uu be a weak solution such that there exists an open internal I⊂(0,∞)I \subset(0,\infty) in which every chain has an upper bound with respect to ⪯\preceq. Then, uu is not Leray-Hopf.

Proof. Again, we prove the contrapositive. Let uu be a Leray-Hopf weak solution. By Theorem 5, we may assume that t→∥u(t)∥t \to\lVert u(t) \rVert is monotone-decreasing on whole of (0,∞)(0,\infty).

Therefore, any open interval II is totally ordered with respect to ⪯\preceq defined on it. Moreover, due to openness, it cannot have an upper bound if viewed as a chain itself.

Indeed, Zorn’s Lemma implies that such an open interval has a maximal element with respect to ⪯\preceq, which is a singular time according to Proposition 9. We observe two consistent flows of results as in the figure below, which illustrates a set-theoretic approach to understanding regularity of weak solutions for turbulence. The upper row in the flow chart is the usual framework in mathematical approach to turbulence. The lower row is the set-theoretic interpretation established in this article.

Figure 1.

References

  1. [1]L. Caffarelli, R. Kohn, and L. Nirenberg, Partial regularity of suitable weak solutions of the Navier-Stokes equations, Comm. Pure Appl. Math. 35 (1982), 771–831.DOI
  2. [2]Danka Lucic and Enrico Pasqualetto, The metric-valued Lebesgue differentiation theorem in measure spaces and its applications, Advances in Operator Theory 8:32 (2023), 1–51.arxiv.org/abs/2111.08526
  3. [3]James C. Robinson, Jose L. Rodrigo, and Witold Sadowski, The Three-Dimensional Navier-Stokes Equations, Cambridge University Press, Cambridge, 2016.
  4. [4]Halsey Royden, Real Analysis, 2nd edition, MacMillan, New York, 1968.

Paper details

vlnc:W_7chwss2r
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 10

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