betaFeedback

WorksMathematics

Rough vector fields with vanishing Lie brackets

Abstract

We consider two vector fields X,Y∈Wloc1,p(Rd)∩L∞X, Y \in W^{1,p}_{loc}(\mathbb R^d) \cap \mathbf L^\infty with bounded divergence. If p≥2p \geq 2, this paper shows that DX Y−DY X=0DX \,Y - DY \, X = 0 almost everywhere if and only if their Regular Lagrangian Flows commute. If instead 1≤p<21 \leq p < 2, we present an example of two divergence free locally Sobolev vector fields in R2\mathbb R^2 with vanishing Lie brackets whose flows do not commute. In the regime 2≤p<∞2 \leq p < \infty this paper also yields a quantitative commutation result not depending on the Lipschitz constant for vector fields with [X,Y]≠0[X, Y] \neq 0, which is new even in the smooth setting.

Introduction

If X∈Wloc1,p(Rd)∩L∞X \in W_{\mathrm{loc}}^{1,p}(\mathbb{R}^{d}) \cap\mathbf{L}^{\infty} is a Sobolev vector field with div⁡X∈L∞\operatorname{div} X \in\mathbf{L}^{\infty}, DiPerna and Lions [4] (and then Ambrosio [1] in the BVBV setting) proved that there is a unique natural flow map ΦXt:Rd→Rd\Phi_X^t:\mathbb{R}^{d} \to\mathbb{R}^{d} such that for all T>0T>0 there is C>0C>0 such that there holds

(ΦXt)♯Ld≤CLdfor all ∣t∣≤T(1)(\Phi_X^t)_{\sharp}\mathcal{L}^{d} \le C\mathcal{L}^{d} \qquad\text{for all } |t| \le T \tag*{(1)}

and for almost every xx, the map t↦ΦXt(x)t \mapsto\Phi_X^t(x) is an absolutely continuous solution to

{ddtΦXt(x)=X(ΦXt(x)),ΦX0(x)=x.(2)\begin{cases} \frac{\mathrm{d}}{\mathrm{d}t}\Phi_X^t(x)=X(\Phi_X^t(x)),\\ \Phi_X^0(x)=x. \tag*{(2)} \end{cases}

The map ΦXt\Phi_X^t is called the Regular Lagrangian flow of the vector field XX. The classical Frobenius theorem states that the (classical) flows ΦXt,ΦYs\Phi_X^t,\Phi_Y^s of two smooth vector fields X,Y:Rd→RdX,Y:\mathbb{R}^{d}\to\mathbb{R}^{d} commute

ΦXt∘ΦYs=ΦYs∘ΦXt,∀ t,s≥0\Phi_X^t \circ\Phi_Y^s = \Phi_Y^s \circ\Phi_X^t,\qquad\forall\ t,s\ge0

if and only if their Lie bracket vanishes, i.e.

[X,Y]:=DX Y−DY X=0.(3)[X,Y] := DX\,Y-DY\,X=0. \tag*{(3)}

The aim of this paper is to extend the classical Frobenius theorem to the setting of Regular Lagrangian Flows.

Let X,Y∈W1,pX,Y \in W^{1,p} be two bounded Sobolev fields, with bounded divergence. If their Regular Lagrangian Flows commute, it is not difficult (see e.g. [2]) to show that [X,Y]=0[X,Y]=0 almost everywhere, therefore this paper and the previous literature on the topic only focus on the opposite implication. Assuming that (3) is satisfied almost everywhere, the following results are known about the commutation of their Regular Lagrangian flows.

  1. A first paper [10] proves that the flows commute when X,YX,Y are locally Lipschitz.

2020 Mathematics Subject Classification. 35F35,37C10,46E35.

Key words and phrases. Sobolev vector fields, Regular Lagrangian flows, Lie brackets.

  1. Later, [2] proved that the flows commute assuming additionally that for every t∈Rt \in\mathbb{R}, the flow map ΦXt\Phi_X^t is weakly Lie differentiable in the direction YY, with locally bounded derivative.

  1. The paper [12] proved the result when one of the vector fields is Lipschitz and the other belongs to W1,pW^{1,p} and admits a unique Regular Lagrangian flow.

  1. Finally [11] proved the commutativity of the flows for bounded, continuous Sobolev vector fields with bounded divergence on R2\mathbb{R}^2.

The main result of this paper is as follows. Here dd is any dimension d≥2d \ge2.

Theorem 1.1. Let X,Y∈Wloc1,p(Rd)∩L∞X,Y \in W_{\mathrm{loc}}^{1,p}(\mathbb{R}^d) \cap L^\infty with div⁡X,div⁡Y∈L∞\operatorname{div} X,\operatorname{div} Y \in L^\infty, and assume that [X,Y]=0[X,Y] = 0. If p≥2p \ge2, then for all t,s>0t,s > 0,

ΦXt∘ΦYs(x)=ΦYs∘ΦXt(x)for a.e. x.(4)\Phi_X^t \circ\Phi_Y^s(x) = \Phi_Y^s \circ\Phi_X^t(x) \qquad\text{for a.e. }x. \tag*{(4)}

If instead 1≤p<21 \le p < 2, there are X,Y∈Wloc1,p(Rd)∩L∞X,Y \in W_{\mathrm{loc}}^{1,p}(\mathbb{R}^d) \cap L^\infty with zero divergence, such that (1.4) fails for xx in an open set with positive measure.

Some comments are in order.

Lie differential and commutation

To understand why a different approach is needed compared to the existing literature, assume that the flow ΦXt\Phi_X^t is Lie-differentiable in the direction of YY, with locally bounded derivative (see [2]). For the purpose of this sketch, assume that there exists a function ρ⃗t∈L∞([0,T]×Rd;Rd)\vec{\rho}_t \in L^\infty([0,T] \times\mathbb{R}^d;\mathbb{R}^d) such that

1h(ΦXt∘ΦYh−ΦXt)⟶ρ⃗tin Lloc1 as h→0+.\frac{1}{h}\left(\Phi_X^t \circ\Phi_Y^h - \Phi_X^t\right) \longrightarrow\vec{\rho}_t \quad\text{in } L^1_{\mathrm{loc}} \text{ as } h \to0^+.

At least formally ρ⃗t\vec{\rho}_t satisfies

∂tρ⃗t=DX(ΦXt)ρ⃗t,ρ⃗0=Y.(5)\partial_t\vec{\rho}_t = DX(\Phi_X^t)\vec{\rho}_t,\qquad\vec{\rho}_0 = Y. \tag*{(5)}

Notice that by applying Gronwall’s estimate along almost every trajectory of XX, the above equation has a unique solution ρ⃗t\vec{\rho}_t in L∞L^\infty. However, due to the assumption [X,Y]=0[X,Y] = 0, it is immediate to check that Y(ΦXt(x))Y(\Phi_X^t(x)) is, in fact, another bounded solution of (1.5), and therefore

ρ⃗t=Y(ΦXt).\vec{\rho}_t = Y(\Phi_X^t).

Therefore

ddsΦXt∘ΦYs(x)=Y(ΦXt∘ΦYs(x))\frac{\mathrm{d}}{\mathrm{d}s}\Phi_X^t \circ\Phi_Y^s(x) = Y(\Phi_X^t \circ\Phi_Y^s(x))

so that s↦ΦXt∘ΦYs(x)s \mapsto\Phi_X^t \circ\Phi_Y^s(x) is a trajectory of the flow of YY for a.e. xx, starting at the point ΦXt(x)\Phi_X^t(x). Since it satisfies (1.1) and (1.2), again by uniqueness of the DiPerna Lions flows one must have that it is equal to ΦYs∘ΦXt(x)\Phi_Y^s \circ\Phi_X^t(x) at a.e. xx.

Results about the differentiability of the Regular Lagrangian Flow of Sobolev fields are available in two dimensions under continuity assumptions [7] (and, classically, for Lipschitz vector fields in general dimension), but it is well known, see [7, 9] that the RLF of Sobolev vector fields need not be differentiable or inherit any Sobolev regularity. In particular, there seems not to be an easy way to fix the previous argument in a general Sobolev setting.

Quantitative commutation

For p>2p > 2, strictly greater than the critical exponent, the proof also provides a quantitative commutation result for vector fields in W1,pW^{1,p}, as follows.

Theorem 1.2. Let p>2p > 2 and let X,Y∈Wloc1,p(Rd)∩L∞X,Y \in W_{\mathrm{loc}}^{1,p}(\mathbb{R}^d) \cap L^\infty with div⁡X,div⁡Y∈L∞\operatorname{div} X,\operatorname{div} Y \in L^\infty. Then

sup⁡∣s∣,∣t∣≤T∫Br∣ΦXt∘ΦYs(x)−ΦYs∘ΦXt(x)∣ dx≲[log⁡(1+1∥[X,Y]∥L1(BR))]−p(6)\sup_{|s|,|t|\le T}\int_{B_r}\left|\Phi_X^t \circ\Phi_Y^s(x)-\Phi_Y^s \circ\Phi_X^t(x)\right|\,\mathrm{d}x \lesssim\left[\log\left(1+\frac{1}{\lVert[X,Y]\rVert_{L^1(B_R)}}\right)\right]^{-p} \tag*{(6)}

where

R=r+T(∥X∥∞+∥Y∥∞).R=r+T(\|X\|_{\infty}+\|Y\|_{\infty}).

The implicit constant depends on d,p,T,rd,p,T,r and the norms of X,YX,Y in BR+1B_{R+1}.

In the critical case X,Y∈W1,2X,Y\in W^{1,2}, one readily deduces the existence of an analogous quantitative result, using the compactness and stability properties of the Regular Lagrangian Flows. However, the uniform modulus that we can obtain when p=2p=2 is not as clean as in the case p>2p>2, as it will be clear from the proof below.

Since X,Y∈L∞X,Y\in L^{\infty}, by a standard localization argument it is easy to see that it is equivalent to prove the commutativity results for vector fields in W1,p(Rd)W^{1,p}(\mathbb{R}^{d}). To avoid technical issues, we will restrict to this class from now on.

The regime p<2p<2

The example of non-commuting flows in the regime p<2p<2 is surprisingly simple, and it is constructed using a double cover of the plane Q:Rx2→Ry2Q:\mathbb{R}_{x}^{2}\to\mathbb{R}_{y}^{2} (see Figure 1). The example was essentially already known from [8] (see also [5, 6]), here we just need to check its Sobolev regularity, which is immediate. Say that, in polar coordinates

Figure 1. Example of two divergence free fields X,Y∈Wloc1,p(R2)X,Y \in W_{\mathrm{loc}}^{1,p}(\mathbb{R}^{2}), p<2p<2, whose flows do not commute. They are constructed as pullbacks of eie_i through a branching map Q:Rx2→Ry2Q:\mathbb{R}_{x}^{2}\to\mathbb{R}_{y}^{2}. For example, in polar coordinates, one can take Q(ρ,θ)=(ρ,2θ)Q(\rho,\theta)=(\rho,2\theta).

Q(ρ,θ)=(ρ,2θ).Q(\rho,\theta)=(\rho,2\theta).

Vector fields with vanishing Lie brackets X,YX,Y producing non-commuting flows can be defined as the pullback of e1=(1,0)e_1=(1,0), e2=(0,1)e_2=(0,1) via the map QQ:

X=(DQ)−1e1,Y=(DQ)−1e2.X=(DQ)^{-1}e_1,\qquad Y=(DQ)^{-1}e_2.

It is easy to see that they satisfy all the conditions of Theorem 1.1 if and only if p<2p<2. Their flows do not commute: the reason is that, while clearly e1,e2e_1,e_2 commute in Ry2\mathbb{R}_{y}^{2}, when they are pulled back via the branching map QQ, the points ΦXt∘ΦYs\Phi_X^t\circ\Phi_Y^s and ΦYs∘ΦXt\Phi_Y^s\circ\Phi_X^t may well end up in different branches of the map QQ in the original space Rx2\mathbb{R}_{x}^{2}. The exponent p=2p=2 arises as a natural threshold also in [6] for the same geometric reasons.

AI use

We started working on the problem in early 2025, but eventually got stuck and moved away from the topic. In September 2026, we were very surprised when GPT-6 Astra suggested (in a single response) the counterexample in the regime p<2p<2. Moreover, this time after extensive interaction with GPT-6 Astra, in late September 2026 we obtained a first proof, completely different from the one presented here, that worked for p>2p > 2 in arbitrary dimension. Finally, in late October 2026, through further interactions with the same model, we obtained the present proof, which we then decided to rework and write up in its current form. The paper was written entirely by the authors.

Short time estimate

We start with some preliminary results, needed for the following.

Maximal functions and truncations

Recall that if f∈W1,p(Rd;Rm)f \in W^{1,p}(\mathbb{R}^{d};\mathbb{R}^{m}) then the maximal function of DfDf is defined as

gf(x):=M(∣Df∣)(x)=sup⁡ρ>01∣Bρ(x)∣∫Bρ(x)∣Df(y)∣ dy.g_f(x) := \mathcal{M}(|Df|)(x) = \sup_{\rho>0} \frac{1}{|B_\rho(x)|} \int_{B_\rho(x)} |Df(y)| \, \mathrm{d}y.

For the following statements, we refer to [13]. If p>1p > 1, one has

∥gf∥Lp≤Cd,p∥f∥W1,p.\|g_f\|_{L^p} \le C_{d,p} \|f\|_{W^{1,p}}.

Then there exists a null set NN such that for all x,y∈Rd∖Nx,y \in\mathbb{R}^{d} \setminus N, there holds

∣f(x)−f(y)∣≤Cd(gf(x)+gf(y))∣x−y∣.|f(x)-f(y)| \le C_d\left(g_f(x)+g_f(y)\right)|x-y|.

On the set

Gλ:={x∈Rd∣gf(x)≤λ2Cd}∖N,G_\lambda:= \left\{x \in\mathbb{R}^{d} \mid g_f(x) \le\frac{\lambda}{2C_d}\right\} \setminus N,

the function ff is λ\lambda-Lipschitz, therefore, by the Kirszbraun Lipschitz extension theorem f∣Gλf|_{G_\lambda} admits a Lipschitz extension fλf_\lambda, with

∥fλ∥∞≤∥f∥∞,Lip⁡fλ≤λ,fλ(x)=f(x)for a.e. x∈Gλ.\|f_\lambda\|_\infty\le\|f\|_\infty,\qquad\operatorname{Lip} f_\lambda\le\lambda,\qquad f_\lambda(x)=f(x)\quad\text{for a.e. }x\in G_\lambda.

Lusin-Lipschitz estimates

It is well known that the Regular Lagrangian Flow of Sobolev vector fields satisfies Lipschitz estimates outside a small set, see [3]. We recall here a version in our setting.

Lemma 2.1. Let p>1p>1 and X∈W1,p(Rd)X \in W^{1,p}(\mathbb{R}^{d}) be a vector field with bounded divergence, and let gXg_X be its maximal function. Then for every λ>0\lambda>0 and 0≤h≤T0\le h\le T there is a set Bλ,hB_{\lambda,h} such that

∣Bλ,h∣≤2Chλ∫{gX≥λ/(2h)}gX(x) dx|B_{\lambda,h}| \le2C \frac{h}{\lambda}\int_{\{g_X\ge\lambda/(2h)\}} g_X(x)\,\mathrm{d}x

and

x↦ΦXh(x)is e2Cdλ-Lipschitz in Rd∖Bλ,h.x\mapsto\Phi_X^h(x)\quad\text{is }e^{2C_d\lambda}\text{-Lipschitz in }\mathbb{R}^{d}\setminus B_{\lambda,h}.

Proof. Up to a mollification argument, we can assume that XX is a smooth vector field. Then we obtain

ddt∣ΦXt(x)−ΦXt(y)∣≤∣X(ΦXt(x))−X(ΦXt(y))∣≤Cd(gX(ΦXt(x))+gX(ΦXt(y)))∣ΦXt(x)−ΦXt(y)∣.\begin{aligned} \frac{\mathrm{d}}{\mathrm{d}t}\left|\Phi_X^t(x)-\Phi_X^t(y)\right| \le\left|X\left(\Phi_X^t(x)\right)-X\left(\Phi_X^t(y)\right)\right| \\ &\le C_d\left(g_X\left(\Phi_X^t(x)\right)+g_X\left(\Phi_X^t(y)\right)\right)\left|\Phi_X^t(x)-\Phi_X^t(y)\right|. \end{aligned}

By Gronwall’s Lemma,

∣ΦXh(x)−ΦXh(y)∣≤∣x−y∣exp⁡(Cd∫0h(gX(ΦXt(x))+gX(ΦXt(y))) dt).\left|\Phi_X^h(x)-\Phi_X^h(y)\right| \le|x-y|\exp\left(C_d\int_0^h\left(g_X\left(\Phi_X^t(x)\right)+g_X\left(\Phi_X^t(y)\right)\right)\,\mathrm{d}t\right).

Define for x∈Rdx\in\mathbb{R}^{d}

Gλ,h(x):=∫0hgX(ΦXt(x)) dtG_{\lambda,h}(x):=\int_0^h g_X\left(\Phi_X^t(x)\right)\,\mathrm{d}t

and define the set

Bλ,h:={x∈Rd∣Gλ,h(x)≥λ}.B_{\lambda,h}:=\left\{x\in\mathbb{R}^{d}\mid G_{\lambda,h}(x)\ge\lambda\right\}.

Then for a.e. x,y∈Rd∖Bλ,hx,y\in\mathbb{R}^{d}\setminus B_{\lambda,h}, there holds

∣ΦXh(x)−ΦXh(y)∣≤e(2Cdλ)∣x−y∣.\left|\Phi_X^h(x)-\Phi_X^h(y)\right|\le e^{(2C d\lambda)}|x-y|.

Moreover,

∣Bλ,h∣≤1λ∫Bλ,hGλ,h(x) dx=1λ∫Bλ,h∫0hgX(ΦXt(x))1{gX(ΦXt(x))≤λ/(2h)}(x) dt dx+1λ∫Bλ,h∫0hgX(ΦXt(x))1{gX(ΦXt(x))>λ/(2h)}(x) dt dx≤12∣Bλ,h∣+Chλ∫RdgX(x)1{gX(x)≥λ/(2h)}(x) dx.\begin{aligned} |B_{\lambda,h}|\le\frac{1}{\lambda}\int_{B_{\lambda,h}}G_{\lambda,h}(x)\,dx \\ &=\frac{1}{\lambda}\int_{B_{\lambda,h}}\int_0^h g_X(\Phi_X^t(x))\mathbf{1}_{\{g_X(\Phi_X^t(x))\le\lambda/(2h)\}}(x)\,dt\,dx \\ &+\frac{1}{\lambda}\int_{B_{\lambda,h}}\int_0^h g_X(\Phi_X^t(x))\mathbf{1}_{\{g_X(\Phi_X^t(x))>\lambda/(2h)\}}(x)\,dt\,dx \\ &\le\frac{1}{2}|B_{\lambda,h}|+\frac{Ch}{\lambda}\int_{\mathbb{R}^{d}}g_X(x)\mathbf{1}_{\{g_X(x)\ge\lambda/(2h)\}}(x)\,dx. \end{aligned}

where we have used (1) to perform the change of variable in the last inequality. Since the first term in the last line can be absorbed in the left hand side, this concludes the proof. □

A short-time estimate

The main goal of this section is to prove the following preliminary proposition, that is a short time estimate on the commutativity of flows when t,s∼ht,s\sim h.

Proposition 2.2. Let X,Y∈W1,p(Rd)∩L∞X,Y\in W^{1,p}(\mathbb{R}^{d})\cap L^\infty be vector fields with bounded divergence. Fix h>0h>0 and r>0r>0. Then for all 0≤t,s≤h0\le t,s\le h there holds

∫Br∣ΦXt∘ΦYs(x)−ΦYs∘ΦXt(x)∣ dx≲h⋅∣{g≥h−1}∩BR∣+h2∥[X,Y]∥L1(BR)(7)\int_{B_r}\left|\Phi_X^t\circ\Phi_Y^s(x)-\Phi_Y^s\circ\Phi_X^t(x)\right|\,dx\lesssim h\cdot\left|\left\{g\ge h^{-1}\right\}\cap B_R\right|+h^2\|[X,Y]\|_{L^1(B_R)} \tag*{(7)}

where

R≡Rr,h:=r+h(∣X∣∞+∣Y∣∞)(8)R\equiv R_{r,h}:=r+h(\lvert X\rvert_\infty+\lvert Y\rvert_\infty) \tag*{(8)}

and g=gX+gYg=g_X+g_Y.

The following is a preliminary estimate for smooth vector fields.

Lemma 2.3. Let X,Y:Rd→RdX,Y:\mathbb{R}^{d}\to\mathbb{R}^{d} be bounded, Lipschitz vector fields with Lipschitz constant at most λ\lambda. Then there holds, for all 0≤t,s≤h:=2Cdλ−10\le t,s\le h:=2C_d\lambda^{-1},

∫Br∣ΦXt∘ΦYs(x)−ΦYs∘ΦXt(x)∣ dx≲dtseλ(s+t)∥[X,Y]∥L1(BRr,h).(9)\int_{B_r}\left|\Phi_X^t\circ\Phi_Y^s(x)-\Phi_Y^s\circ\Phi_X^t(x)\right|\,dx\lesssim_d ts e^{\lambda(s+t)}\|[X,Y]\|_{L^1(B_{R_{r,h}})}. \tag*{(9)}

where the implicit constant is a dimensional constant.

Proof. Without loss of generality we prove the result for t,s≥0t,s\ge0. By a standard mollification argument, it is sufficient to prove the result for smooth vector fields. First consider the quantity

Et(z):=DΦXt(z)Y(z)−Y(ΦXt(z)),t≥0,z∈RdE^t(z):=D\Phi_X^t(z)Y(z)-Y(\Phi_X^t(z)),\qquad t\ge0,\quad z\in\mathbb{R}^{d}

Differentiating with respect to tt, we obtain

∂tEt(z)=DX(ΦXt(z))Et(z)+[X,Y](ΦXt(z)),E0(z)=0.\partial_tE^t(z)=DX(\Phi_X^t(z))E^t(z)+[X,Y](\Phi_X^t(z)),\qquad E_0(z)=0.

Therefore, by Gronwall’s Lemma we obtain

∣Et(z)∣≤eλt∫0t∣[X,Y](ΦXp(z))∣ dp.(10)|E^t(z)|\le e^{\lambda t}\int_0^t|[X,Y](\Phi_X^p(z))|\,dp. \tag*{(10)}

Next, consider the function

[0,s]∋u↦Ct,u(x):=ΦXt∘ΦYu(x)−ΦYu∘ΦXt(x).[0,s]\ni u\mapsto C^{t,u}(x):=\Phi_X^t\circ\Phi_Y^u(x)-\Phi_Y^u\circ\Phi_X^t(x).

Its derivative with respect to uu is

∂uCt,u(x)=DΦXt(ΦYu(x)) Y(ΦYu(x))−Y(ΦYu∘ΦXt(x))=(DΦXt(ΦYu(x)) Y(ΦYu(x))−Y(ΦXt∘ΦYu(x)))+(Y(ΦXt∘ΦYu(x))−Y(ΦYu∘ΦXt(x))).\begin{aligned} \partial_u C^{t,u}(x)=D\Phi_X^t(\Phi_Y^u(x))\,Y(\Phi_Y^u(x))-Y(\Phi_Y^u\circ\Phi_X^t(x)) \\ &=\left(D\Phi_X^t(\Phi_Y^u(x))\,Y(\Phi_Y^u(x))-Y(\Phi_X^t\circ\Phi_Y^u(x))\right) \\ &+\left(Y(\Phi_X^t\circ\Phi_Y^u(x))-Y(\Phi_Y^u\circ\Phi_X^t(x))\right). \end{aligned}

Therefore, by triangular inequality, using (10)

∣∂uCt,u(x)∣≤∣Et(ΦYu(x))∣+λ∣Ct,u(x)∣≤eλt∫0t∣[X,Y](ΦXp∘ΦYu(x))∣ dp+λ∣Ct,u(x)∣.\begin{aligned} |\partial_u C^{t,u}(x)|\leq|\mathbb{E}^t(\Phi_Y^u(x))|+\lambda|C^{t,u}(x)| \\ &\leq e^{\lambda t}\int_0^t|[X,Y](\Phi_X^p\circ\Phi_Y^u(x))|\,\mathrm{d}p+\lambda|C^{t,u}(x)|. \end{aligned}

Therefore again by Gronwall’s Lemma we obtain

∣ΦXt∘ΦYs−ΦYs∘ΦXt(x)∣≤eλ(s+t)∫0s∫0t∣[X,Y](ΦXp∘ΦYu(x))∣ dp du|\Phi_X^t\circ\Phi_Y^s-\Phi_Y^s\circ\Phi_X^t(x)|\leq e^{\lambda(s+t)}\int_0^s\int_0^t|[X,Y](\Phi_X^p\circ\Phi_Y^u(x))|\,\mathrm{d}p\,\mathrm{d}u

A final integration in xx yields

∫Br∣ΦXt∘ΦYs−ΦYs∘ΦXt(x)∣ dx≤eλ(s+t)∫0s∫0t(∫Br∣[X,Y](ΦXp∘ΦYu(x))∣ dx) dp du≤Ctseλ(s+t)∫BR∣[X,Y](x)∣ dx\begin{aligned} \int_{B_r}|\Phi_X^t\circ\Phi_Y^s-\Phi_Y^s\circ\Phi_X^t(x)|\,\mathrm{d}x\leq e^{\lambda(s+t)}\int_0^s\int_0^t\left(\int_{B_r}|[X,Y](\Phi_X^p\circ\Phi_Y^u(x))|\,\mathrm{d}x\right)\,\mathrm{d}p\,\mathrm{d}u \\ &\leq Ct s e^{\lambda(s+t)}\int_{B_R}|[X,Y](x)|\,\mathrm{d}x \end{aligned}

where in the last inequality we use the compression bounds on the flows of X,YX,Y for small times 0≤t,s≤2Cdλ−10\leq t,s\leq2Cd\lambda^{-1}, which imply

(ΦXp)#Ld,(ΦYu)#Ld≤CLd(\Phi_X^p)_{\#}\mathcal{L}^d,\quad(\Phi_Y^u)_{\#}\mathcal{L}^d\leq C\mathcal{L}^d

where CC is a dimensional constant.

Lemma 2.4. If X∈W1,p(Rd)∩L∞X\in W^{1,p}(\mathbb{R}^d)\cap L^\infty has bounded divergence, and XλX_\lambda is its λ\lambda-Lipschitz truncation, then there holds for all 0≤t≤T0\leq t\leq T

∫Br∣ΦXt(x)−ΦXλt(x)∣ dx≤Cteλt∣{X≠Xλ}∩Br+t∣X∣∞∣.\int_{B_r}|\Phi_X^t(x)-\Phi_{X_\lambda}^t(x)|\,\mathrm{d}x\leq Ct e^{\lambda t}\left|\{X\neq X_\lambda\}\cap B_{r+t|X|_\infty}\right|.

Proof. We have

∂t(ΦXt(x)−ΦXλt(x))=(X(ΦXt(x))−Xλ(ΦXt(x)))+(Xλ(ΦXt(x))−Xλ(ΦXλt(x)))\begin{aligned} \partial_t\left(\Phi_X^t(x)-\Phi_{X_\lambda}^t(x)\right)=\left(X(\Phi_X^t(x))-X_\lambda(\Phi_X^t(x))\right) \\ &+\left(X_\lambda(\Phi_X^t(x))-X_\lambda(\Phi_{X_\lambda}^t(x))\right) \end{aligned}

By estimating the second term using the Lipschitz regularity of XλX_\lambda, again by Gronwall’s Lemma we obtain

∣ΦXt(x)−ΦXλt(x)∣≤eλt∫0t∣X(ΦXs(x))−Xλ(ΦXs(x))∣ ds|\Phi_X^t(x)-\Phi_{X_\lambda}^t(x)|\leq e^{\lambda t}\int_0^t|X(\Phi_X^s(x))-X_\lambda(\Phi_X^s(x))|\,\mathrm{d}s

Integrating in xx

∫Br∣ΦXt(x)−ΦXλt(x)∣ dx≤Cteλt∫Br∣X(x)−Xλ(x)∣ dx≤∥X∥∞Cteλt∣{X≠Xλ}∩BR∣\begin{aligned} \int_{B_r}|\Phi_X^t(x)-\Phi_{X_\lambda}^t(x)|\,\mathrm{d}x\leq Ct e^{\lambda t}\int_{B_r}|X(x)-X_\lambda(x)|\,\mathrm{d}x \\ &\leq\|X\|_\infty Ct e^{\lambda t}|\{X\neq X_\lambda\}\cap B_R| \end{aligned}

where in the first inequality we used the compression bounds (1) for the Regular Lagrangian flow of XX.

Proof of Proposition 2.2. We set

λ:=2Cd/h\lambda:= 2C_d/h

and we consider the truncated fields Xλ,YλX_\lambda,Y_\lambda. We can write

∫Br∣ΦXt∘ΦYs(x)−ΦYs∘ΦXt(x)∣ dx≤∫Br∣ΦXλt∘ΦYλs(x)−ΦYλs∘ΦXλt(x)∣ dx+∫Br∣ΦXλt∘ΦYs(x)−ΦXt∘ΦYλs(x)∣ dx+∫Br∣ΦYλs∘ΦXt(x)−ΦYs∘ΦXλt(x)∣ dx.(11)\begin{aligned} \int_{B_r}\left|\Phi_X^t\circ\Phi_Y^s(x)-\Phi_Y^s\circ\Phi_X^t(x)\right|\,\mathrm{d}x \leq\int_{B_r}\left|\Phi_{X_\lambda}^t\circ\Phi_{Y_\lambda}^s(x)-\Phi_{Y_\lambda}^s\circ\Phi_{X_\lambda}^t(x)\right|\,\mathrm{d}x \\ &+\int_{B_r}\left|\Phi_{X_\lambda}^t\circ\Phi_Y^s(x)-\Phi_X^t\circ\Phi_{Y_\lambda}^s(x)\right|\,\mathrm{d}x \\ &+\int_{B_r}\left|\Phi_{Y_\lambda}^s\circ\Phi_X^t(x)-\Phi_Y^s\circ\Phi_{X_\lambda}^t(x)\right|\,\mathrm{d}x. \tag*{(11)} \end{aligned}

Notice that by applying Lemma 2.4 twice, there holds

∫Br∣ΦXλt∘ΦYλs(x)−ΦXλt∘ΦYs(x)∣ dx≲∫Br∣ΦXλt∘ΦYλs(x)−ΦXt∘ΦYλs(x)∣ dx+∫Br∣ΦXλt∘ΦYs(x)−ΦXt∘ΦYs(x)∣ dx\begin{aligned} \int_{B_r}\left|\Phi_{X_\lambda}^t\circ\Phi_{Y_\lambda}^s(x)-\Phi_{X_\lambda}^t\circ\Phi_Y^s(x)\right|\,\mathrm{d}x \lesssim\int_{B_r}\left|\Phi_{X_\lambda}^t\circ\Phi_{Y_\lambda}^s(x)-\Phi_X^t\circ\Phi_{Y_\lambda}^s(x)\right|\,\mathrm{d}x \\ &+\int_{B_r}\left|\Phi_{X_\lambda}^t\circ\Phi_Y^s(x)-\Phi_X^t\circ\Phi_Y^s(x)\right|\,\mathrm{d}x \end{aligned}
≲eλt∫Br∣ΦYλs(x)−ΦYs(x)∣ dx+teλt∣{X≠Xλ}∩BR∣\lesssim e^{\lambda t}\int_{B_r}\left|\Phi_{Y_\lambda}^s(x)-\Phi_Y^s(x)\right|\,\mathrm{d}x+te^{\lambda t}\left|\{X\ne X_\lambda\}\cap B_R\right|
≲seλteλs∣{Y≠Yλ}∩BR∣+teλt∣{X≠Xλ}∩BR∣.\lesssim se^{\lambda t}e^{\lambda s}\left|\{Y\ne Y_\lambda\}\cap B_R\right|+te^{\lambda t}\left|\{X\ne X_\lambda\}\cap B_R\right|.

The third term in (11) can be estimated in an entirely symmetrical way, while for the first one can use Lemma 2.3. Combining all these estimates, and recalling the properties of the Lipschitz truncation of Section 2.1, one gets

∫Br∣ΦXt∘ΦYs(x)−ΦYs∘ΦXt(x)∣ dx≤h2∫BR∣[Xλ,Yλ](x)∣ dx+h∣BR∩{g≥h−1}∣\int_{B_r}\left|\Phi_X^t\circ\Phi_Y^s(x)-\Phi_Y^s\circ\Phi_X^t(x)\right|\,\mathrm{d}x \leq h^2\int_{B_R}\left|[X_\lambda,Y_\lambda](x)\right|\,\mathrm{d}x+h\left|B_R\cap\{g\geq h^{-1}\}\right|
≲h2∫BR∣[X,Y]∣ dx+h∣BR∩{g≥h−1}∣.\lesssim h^2\int_{B_R}\left|[X,Y]\right|\,\mathrm{d}x+h\left|B_R\cap\{g\geq h^{-1}\}\right|.

where we used that ∣DXλ∣,∣DYλ∣≤λ=2Cd/h|DX_\lambda|,|DY_\lambda|\leq\lambda=2C_d/h to pass from the first to the second line and moreover that, on the good set {g≤h−1}\{g\leq h^{-1}\}, by locality of the weak gradient one has DX=DXλDX=DX_\lambda and DY=DYλDY=DY_\lambda almost everywhere.

A quantitative commutation estimate

In this section we prove the first part of Theorem 1.1, i.e. that the flows commute whenever p≥2p\geq2.

The following is the key proposition: the initial commutator is split into 4k4^k smaller commutators, up to removing a set with controlled measure. Here we recall that gg is the sum of the maximal functions of the derivatives of X,YX,Y:

g=gX+gYg=g_X+g_Y

as defined in Section 2.1.

Proposition 3.1. Let X,Y∈W1,p(Rd)∩L∞X,Y\in W^{1,p}(\mathbb{R}^d)\cap L^\infty, with p>1p>1 and bounded divergence. Let 0≤s,t≤T0\leq s,t\leq T, and let {λj}j∈N\{\lambda_j\}_{j\in\mathbb{N}} be a sequence of positive numbers. Then, for every

r>0r > 0 and k>0k > 0,

∫Br∣ΦXt∘ΦYs(x)−ΦYs∘ΦXt(x)∣ dx≲∑j=1k2jλj∫{g>2jλj/(2T)}g(x) dx+4k(∏j=1kecdλj)∫BRk∣ΦXt/2k∘ΦYs/2k(x)−ΦYs/2k∘ΦXt/2k(x)∣ dx.(12)\int_{B_r}\left|\Phi_X^t \circ\Phi_Y^s(x)-\Phi_Y^s \circ\Phi_X^t(x)\right|\,\mathrm{d}x \lesssim\sum_{j=1}^{k}\frac{2^j}{\lambda_j}\int_{\{g>2^j\lambda_j/(2T)\}}g(x)\,\mathrm{d}x +4^k\left(\prod_{j=1}^{k}e^{c_d\lambda_j}\right)\int_{B_{R_k}}\left|\Phi_X^{t/2^k}\circ\Phi_Y^{s/2^k}(x)-\Phi_Y^{s/2^k}\circ\Phi_X^{t/2^k}(x)\right|\,\mathrm{d}x. \tag*{(12)}

where here

Rk=r+(1−2−k)T(∣X∣∞+∣Y∣∞).R_k=r+(1-2^{-k})T(\lvert X\rvert_\infty+\lvert Y\rvert_\infty).

and cdc_d is a dimensional constant, and the implicit constant depends on r,T,dr,T,d.

Proof of Proposition 3.1. 1. For convenience of notation, let us introduce, for all j∈Nj\in\mathbb{N},

Aj:=ΦXt/2j,Bj:=ΦYs/2j,A_j:=\Phi_X^{t/2^j},\qquad B_j:=\Phi_Y^{s/2^j},
Γj:=Aj∘Bj−Bj∘Aj.\Gamma_j:=A_j\circ B_j-B_j\circ A_j.

Notice the recursive relation

Γj=Aj+1∘Γj+1∘Bj+1+Aj+1∘Bj+1∘Γj+1+Γj+1∘Bj+1∘Aj+1+Bj+1∘Γj+1∘Aj+1.(13)\begin{aligned} \Gamma_j={}&A_{j+1}\circ\Gamma_{j+1}\circ B_{j+1}\\ &+A_{j+1}\circ B_{j+1}\circ\Gamma_{j+1}\\ &+\Gamma_{j+1}\circ B_{j+1}\circ A_{j+1}\\ &+B_{j+1}\circ\Gamma_{j+1}\circ A_{j+1}. \tag*{(13)} \end{aligned}

Here we use the convention that for two maps D1,D2:Rd→RdD_1,D_2:\mathbb{R}^d\to\mathbb{R}^d,

D1∘Γj∘D2≡D1∘Aj∘Bj∘D2−D1∘Bj∘Aj∘D2.D_1\circ\Gamma_j\circ D_2\equiv D_1\circ A_j\circ B_j\circ D_2-D_1\circ B_j\circ A_j\circ D_2.

Introduce the set of indices Ik:={1,2,3,4}k\mathcal{I}^k:=\{1,2,3,4\}^k. An element α∈{1,2,3,4}k\alpha\in\{1,2,3,4\}^k is a vector of the form

α=(α1,…,αk),αi∈{1,2,3,4}.\alpha=(\alpha^1,\ldots,\alpha^k),\qquad\alpha^i\in\{1,2,3,4\}.

Expanding the first commutator Γ0\Gamma_0 recursively for k≥1k\geq1 iterations, we therefore obtain for every k≥1k\geq1 that

Γ0=∑α∈Ik(Lα11∘…∘Lαkk)∘Γk∘(Rαkk∘…∘Rα11).(14)\Gamma_0=\sum_{\alpha\in\mathcal{I}^k}\left(\mathcal{L}_{\alpha^1}^1\circ\ldots\circ\mathcal{L}_{\alpha^k}^k\right)\circ\Gamma_k\circ\left(\mathcal{R}_{\alpha^k}^k\circ\ldots\circ\mathcal{R}_{\alpha^1}^1\right). \tag*{(14)}

In particular, at step kk, we decomposed Γ0\Gamma_0 into the sum of 4k4^k commutators Γk\Gamma_k, left and right composed with some operators. Here Lαjj\mathcal{L}_{\alpha^j}^j is one of the four elements appearing in the left compositions in (13), more precisely:

Lαjj∈{Aj,Aj∘Bj,Id⁡,Bj}.(15)\mathcal{L}_{\alpha^j}^j\in\{A_j,A_j\circ B_j,\operatorname{Id},B_j\}. \tag*{(15)}

In an entirely similar way

Rαjj∈{Bj,Id⁡,Bj∘Aj,Aj}.(16)\mathcal{R}_{\alpha^j}^j\in\{B_j,\operatorname{Id},B_j\circ A_j,A_j\}. \tag*{(16)}

Let us name

L~αk:=Lα11∘…∘Lαkk,R~αk:=Rαkk∘…∘Rα11.\widetilde{\mathcal{L}}_\alpha^k:=\mathcal{L}_{\alpha^1}^1\circ\ldots\circ\mathcal{L}_{\alpha^k}^k,\qquad \widetilde{\mathcal{R}}_\alpha^k:=\mathcal{R}_{\alpha^k}^k\circ\ldots\circ\mathcal{R}_{\alpha^1}^1.

Our aim is to estimate at each step kk the commutator Γk\Gamma_k with the commutator Γk+1\Gamma_{k+1}, up to removing a small set where the new four left compositions Lpk+1\mathcal{L}_p^{k+1}, p=1,2,3,4p=1,2,3,4 are not eλie^{\lambda_i}-Lipschitz.

Notice that for every j≥1j\geq1, Lpj\mathcal{L}_p^j is a composition of the flows of X,YX,Y for a total time of order 2−j2^{-j}, as it immediately follows from (16). Recall that g=gX+gYg=g_X+g_Y is the sum of the maximal functions of DXDX, DYDY, which are in Llocp\mathbf{L}^p_{\mathrm{loc}}. By Lemma 2.1, for all j≥0j \ge0 and p∈{1,2,3,4}p \in\{1,2,3,4\}, there exist sets Bpj\mathcal{B}_p^j such that

∣Bpj∣≲1λj2j∫{g≥λj2j/(2T)}g(x) dx.\lvert\mathcal{B}_p^j \rvert\lesssim\frac{1}{\lambda_j 2^j}\int_{\{g\geq\lambda_j 2^j/(2T)\}} g(x)\,\mathrm{d}x.

and

Lpj is ecdλj-Lipschitz on Rd∖Bpj.(17)\mathcal{L}_p^j\text{ is }e^{c d\lambda_j}\text{-Lipschitz on }\mathbb{R}^d\setminus\mathcal{B}_p^j. \tag*{(17)}

Define the bad set:

Bj:=⋃α∈Ij⋃p=14{x∈Rd∣Aj∘Bj∘R~αj(x)∈Bpj}∪⋃α∈Ij⋃p=14{x∈Rd∣Bj∘Aj∘R~αj(x)∈Bpj}.\mathcal{B}_j:=\bigcup_{\alpha\in\mathcal{I}^j}\bigcup_{p=1}^{4}\left\{x\in\mathbb{R}^d\mid A_j\circ B_j\circ\widetilde{\mathcal{R}}_\alpha^j(x)\in\mathcal{B}_p^j\right\} \cup\bigcup_{\alpha\in\mathcal{I}^j}\bigcup_{p=1}^{4}\left\{x\in\mathbb{R}^d\mid B_j\circ A_j\circ\widetilde{\mathcal{R}}_\alpha^j(x)\in\mathcal{B}_p^j\right\}.

Therefore we estimate

∣Bj∣≤2⋅4⋅4j⋅1λj2j∫{g≥λj2j/(2T)}g(x) dx.(18)\lvert\mathcal{B}_j\rvert\leq2\cdot4\cdot4^j\cdot\frac{1}{\lambda_j2^j}\int_{\{g\geq\lambda_j2^j/(2T)\}}g(x)\,\mathrm{d}x. \tag*{(18)}

The good set at stage kk is then

Gk:=Rd∖⋃j=1kBj.\mathcal{G}_k:=\mathbb{R}^d\setminus\bigcup_{j=1}^{k}\mathcal{B}_j.

Therefore, in the good region Gk\mathcal{G}_k we have that every appearing left factor Lpj\mathcal{L}_p^j is eλje^{\lambda_j}-Lipschitz on the points given by the right composition.

2. We want to show that for every k′≤kk' \leq k, there holds

∫Gk∩Br∣Γ0(x)∣ dx≤∑α∈Ik′(∏j=1k′ecdλj)∫Gk∩Br∣Γk′∘R~αk′∣ dx.(19)\int_{\mathcal{G}_k\cap B_r}\lvert\Gamma_0(x)\rvert\,\mathrm{d}x \leq\sum_{\alpha\in\mathcal{I}^{k'}}\left(\prod_{j=1}^{k'}e^{c d\lambda_j}\right)\int_{\mathcal{G}_k\cap B_r}\lvert\Gamma_{k'}\circ\widetilde{\mathcal{R}}_\alpha^{k'}\rvert\,\mathrm{d}x. \tag*{(19)}

This can be done, for example, by an easy induction. For k′=0k'=0, the result is trivial (here for k=0k=0, we set Rα0=Id⁡\mathcal{R}_\alpha^0=\operatorname{Id}).

Assume to have proved (19) up to k′∈Nk'\in\mathbb{N}, where 1≤k′<k1\leq k'<k. Decomposing Γk′\Gamma_{k'} as in (13), we obtain

∑α∈Ik′(∏j=1k′ecdλj)∫Gk∩Br∣Γk′∘R~αk′∣ dx≤∑p=14∑α∈Ik′(∏j=1k′ecdλj)∫Gk∩Br∣Lpk′+1∘Γk′+1∘Rpk′+1∘R~αk′∣ dx=∑α∈Ik′+1(∏j=1k′ecdλj)∫Gk∩Br∣Lαk′+1k′+1∘Γk′+1∘R~αk′+1∣ dx≤∑α∈Ik′+1(∏j=1k′+1ecdλj)∫Gk∩Br∣Γk′+1∘R~αk′+1∣ dx.(20)\begin{aligned} \sum_{\alpha\in\mathcal{I}^{k'}}\left(\prod_{j=1}^{k'}e^{c d\lambda_j}\right)\int_{\mathcal{G}_k\cap B_r}\lvert\Gamma_{k'}\circ\widetilde{\mathcal{R}}_\alpha^{k'}\rvert\,\mathrm{d}x \\ &\leq\sum_{p=1}^{4}\sum_{\alpha\in\mathcal{I}^{k'}}\left(\prod_{j=1}^{k'}e^{c d\lambda_j}\right)\int_{\mathcal{G}_k\cap B_r}\lvert\mathcal{L}_p^{k'+1}\circ\Gamma_{k'+1}\circ\mathcal{R}_p^{k'+1}\circ\widetilde{\mathcal{R}}_\alpha^{k'}\rvert\,\mathrm{d}x \\ &=\sum_{\alpha\in\mathcal{I}^{k'+1}}\left(\prod_{j=1}^{k'}e^{c d\lambda_j}\right)\int_{\mathcal{G}_k\cap B_r}\lvert\mathcal{L}_{\alpha_{k'+1}}^{k'+1}\circ\Gamma_{k'+1}\circ\widetilde{\mathcal{R}}_\alpha^{k'+1}\rvert\,\mathrm{d}x \\ &\leq\sum_{\alpha\in\mathcal{I}^{k'+1}}\left(\prod_{j=1}^{k'+1}e^{c d\lambda_j}\right)\int_{\mathcal{G}_k\cap B_r}\lvert\Gamma_{k'+1}\circ\widetilde{\mathcal{R}}_\alpha^{k'+1}\rvert\,\mathrm{d}x. \tag*{(20)} \end{aligned}

Here in the equality of the third line we used the fact that if α∈Ik′\alpha\in\mathcal{I}^{k'} and p=αk′+1p=\alpha^{k'+1}, then α′:=(α,p)∈Ik′+1\alpha':=(\alpha,p)\in\mathcal{I}^{k'+1}, and by definition

Rpk′+1∘R~αk′=R~α′k′+1\mathcal{R}_p^{k'+1}\circ\widetilde{\mathcal{R}}_\alpha^{k'}=\widetilde{\mathcal{R}}_{\alpha'}^{k'+1}

In the last inequality, we used the fact that Lαk′+1k′+1\mathcal{L}_{\alpha_{k'+1}}^{k'+1} is calculated only on pair of points where it is eλk′+1e^{\lambda_{k'+1}}-Lipschitz thanks to the definition of the good set GkG_k. This proves the inductive step, so that (19) holds in particular for k′=kk'=k, for every k≥1k \ge1.

Next, let k∈Nk \in\mathbb{N}, apply (19) with k′=kk'=k, and write

∫Br∣Γ0(x)∣ dx≲∑j=1k∣Bj∣+∑α∈Tk(∏j=1kecdλj)∫Gk∩Br∣Γk∘R~αk∣ dx≲∑j=1k2jλj∫{g≥λj2j/(2T)}g(x) dx+4k(∏j=1kecdλj)∫BRk∣Γk(x)∣ dx.(21)\begin{aligned} \int_{B_r} |\Gamma_0(x)|\,\mathrm{d}x &\lesssim\sum_{j=1}^{k} |B_j|+\sum_{\alpha\in\mathcal{T}^k}\left(\prod_{j=1}^{k}e^{c_d\lambda_j}\right)\int_{G_k\cap B_r}|\Gamma_k\circ\widetilde{\mathcal{R}}_\alpha^k|\,\mathrm{d}x\\ &\lesssim\sum_{j=1}^{k}\frac{2^j}{\lambda_j}\int_{\{g\geq\lambda_j2^j/(2T)\}}g(x)\,\mathrm{d}x+4^k\left(\prod_{j=1}^{k}e^{c_d\lambda_j}\right)\int_{B_{R_k}}|\Gamma_k(x)|\,\mathrm{d}x. \tag*{(21)} \end{aligned}

Here in the last step we used that every right factor R~αk\widetilde{\mathcal{R}}_\alpha^k has uniformly bounded compression, since it is a composition of flows of time at most 2T2T, and that the points R~αk\widetilde{\mathcal{R}}_\alpha^k are contained in the ball BRkB_{R_k}.

We are now in the position of proving the main theorem for vector fields in W1,2W^{1,2}.

Theorem 3.2. Let X,Y∈W1,2(Rd)∩L∞X,Y\in W^{1,2}(\mathbb{R}^d)\cap L^\infty with bounded divergence, and assume that [X,Y]=0[X,Y]=0. Then for all s,ts,t, there holds

ΦXt∘ΦYs(x)−ΦYs∘ΦXt(x)=0for a.e. x∈Rd.(22)\Phi_X^t\circ\Phi_Y^s(x)-\Phi_Y^s\circ\Phi_X^t(x)=0\qquad\text{for a.e. }x\in\mathbb{R}^d. \tag*{(22)}

Proof. Let r>0r>0 be fixed and take 0≤s,t≤T=10\leq s,t\leq T=1. From Proposition 3.1, we deduce

∫Br∣ΦXt∘ΦYs(x)−ΦYs∘ΦXt(x)∣ dx≲∑j=1k2jλj∫{g>2jλj/(2T)}g(x) dx+4k(∏j=1kecdλj)∫BRk∣ΦXt/2k∘ΦYs/2k(x)−ΦYs/2k∘ΦXt/2k(x)∣ dx.(23)\begin{aligned} \int_{B_r}\left|\Phi_X^t\circ\Phi_Y^s(x)-\Phi_Y^s\circ\Phi_X^t(x)\right|\,\mathrm{d}x &\lesssim\sum_{j=1}^{k}\frac{2^j}{\lambda_j}\int_{\{g>2^j\lambda_j/(2T)\}}g(x)\,\mathrm{d}x\\ &\quad+4^k\left(\prod_{j=1}^{k}e^{c_d\lambda_j}\right)\int_{B_{R_k}}\left|\Phi_X^{t/2^k}\circ\Phi_Y^{s/2^k}(x)-\Phi_Y^{s/2^k}\circ\Phi_X^{t/2^k}(x)\right|\,\mathrm{d}x. \tag*{(23)} \end{aligned}

Let η>0\eta>0 be fixed but sufficiently small, and M>0M>0. Define

λj:=η+M2−j\lambda_j:=\eta+M2^{-j}

where η\eta is fixed and sufficiently small to be chosen later. The first term is then bounded by

∑j=1k2jλj∫{g>2jλj/(2T)}g(x) dx=∫{g≥M/(2T)}g(x)∑j=1k2jλj1{g>2jλj/(2T)}(x) dx≤1η∫{g≥M/(2T)}g(x)∑j=1⌊log⁡2(2Tg/η)⌋2j dx≤4Tη2∫{g≥M/(2T)}g2(x) dx.(24)\begin{aligned} \sum_{j=1}^{k}\frac{2^j}{\lambda_j}\int_{\{g>2^j\lambda_j/(2T)\}}g(x)\,\mathrm{d}x &=\int_{\{g\geq M/(2T)\}}g(x)\sum_{j=1}^{k}\frac{2^j}{\lambda_j}\mathbf{1}_{\{g>2^j\lambda_j/(2T)\}}(x)\,\mathrm{d}x\\ &\leq\frac{1}{\eta}\int_{\{g\geq M/(2T)\}}g(x)\sum_{j=1}^{\lfloor\log_2(2Tg/\eta)\rfloor}2^j\,\mathrm{d}x\\ &\leq\frac{4T}{\eta^2}\int_{\{g\geq M/(2T)\}}g^2(x)\,\mathrm{d}x. \tag*{(24)} \end{aligned}

For the second term, we use Proposition 2.2, so that we obtain

4k(∏j=1kecdλj)∫BRk∣ΦXt2k∘ΦYs2k(x)−ΦYs2k∘ΦXt2k(x)∣ dx≲4kecd(kη+2M)2−k∣{g≥2k/T}∣≲e2cdMecdkη2−k∫{g≥2k/T}q2(x) dx(25)4^k\left(\prod_{j=1}^{k}e^{c_d\lambda_j}\right)\int_{B_{R_k}}\left|\Phi_X^{\frac{t}{2^k}}\circ\Phi_Y^{\frac{s}{2^k}}(x)-\Phi_Y^{\frac{s}{2^k}}\circ\Phi_X^{\frac{t}{2^k}}(x)\right|\,dx \lesssim4^k e^{c_d(k\eta+2M)}2^{-k}\left|\left\{g\geq2^k/T\right\}\right| \lesssim e^{2c_dM}e^{c_dk\eta}2^{-k}\int_{\{g\geq2^k/T\}}q^2(x)\,dx \tag*{(25)}

If η\eta is sufficiently small (η≤log⁡(2)/cd\eta\leq\log(2)/c_d), the right hand side goes to zero as k→+∞k\to+\infty and MM is fixed. Then, we let MM go to infinity, so that the right hand side of (24) goes to zero as well. This proves the result. □

For p>2p>2, there is a clean quantitative version using a different choice of the sequence λi\lambda_i.

Theorem 3.3. Let X,Y∈W1,p(Rd)∩L∞X,Y\in W^{1,p}(\mathbb{R}^d)\cap L^\infty with 2<p<∞2<p<\infty, with bounded divergence. Then for all 0≤s,t≤T0\leq s,t\leq T, there holds

∫Br∣ΦXt∘ΦYs(x)−ΦYs∘ΦXt(x)∣ dx≲[log⁡(1+1∥[X,Y]∥L1(BR))]−p.(26)\int_{B_r}\left|\Phi_X^t\circ\Phi_Y^s(x)-\Phi_Y^s\circ\Phi_X^t(x)\right|\,dx\lesssim\left[\log\left(1+\frac{1}{\|[X,Y]\|_{L^1(B_R)}}\right)\right]^{-p}. \tag*{(26)}

where RR is as in Proposition 3.1.

Proof. We apply the same argument as in the proof of the theorem above. This time, choose

λj=λ2−αj,α=p−22p.\lambda_j=\lambda2^{-\alpha j},\qquad\alpha=\frac{p-2}{2p}.

From (23), we can now estimate the first term in the right hand side as

∑j=1k2jλj∫{g>2jλj/(2T)}g(x) dx≤1λp(∑j=1k2j(1−12))∫Rdgp(x) dx≤Cpλp∫Rdgp(x) dx(27)\sum_{j=1}^{k}\frac{2^j}{\lambda_j}\int_{\{g>2^j\lambda_j/(2T)\}}g(x)\,dx \leq\frac{1}{\lambda^p}\left(\sum_{j=1}^{k}2^{j(1-\frac{1}{2})}\right)\int_{\mathbb{R}^d}g^p(x)\,dx \leq\frac{C_p}{\lambda^p}\int_{\mathbb{R}^d}g^p(x)\,dx \tag*{(27)}

where CpC_p is a constant depending on pp. The second term in (23) is now estimated by Proposition 2.2 as

4k(∏j=1kecdλj)∫BRk∣ΦXt2k∘ΦYs2k(x)−ΦYs2k∘ΦXt2k(x)∣ dx≲4keC~pλ(2−k∣{g≥2k}∣+2−2k∥[X,Y]∥L1(BR))≲eC~pλ2−(p−1)k∫Rdgp(x) dx+eC~pλ∥[X,Y]∥L1(BR).(28)4^k\left(\prod_{j=1}^{k}e^{c_d\lambda_j}\right)\int_{B_{R_k}}\left|\Phi_X^{\frac{t}{2^k}}\circ\Phi_Y^{\frac{s}{2^k}}(x)-\Phi_Y^{\frac{s}{2^k}}\circ\Phi_X^{\frac{t}{2^k}}(x)\right|\,dx \lesssim4^k e^{\widetilde C_p\lambda}\left(2^{-k}\left|\{g\geq2^k\}\right|+2^{-2k}\|[X,Y]\|_{L^1(B_R)}\right) \lesssim e^{\widetilde C_p\lambda}2^{-(p-1)k}\int_{\mathbb{R}^d}g^p(x)\,dx+e^{\widetilde C_p\lambda}\|[X,Y]\|_{L^1(B_R)}. \tag*{(28)}

Then, send kk to infinity. Since p>2p>2 the first term in (28) vanishes, and we get

∫Br∣ΦXt∘ΦYs(x)−ΦYs∘ΦXt(x)∣ dx≤Cpλp∫Rdgp(x) dx+eC~pλ∥[X,Y]∥L1(BR).(29)\int_{B_r}\left|\Phi_X^t\circ\Phi_Y^s(x)-\Phi_Y^s\circ\Phi_X^t(x)\right|\,dx\leq\frac{C_p}{\lambda^p}\int_{\mathbb{R}^d}g^p(x)\,dx+e^{\widetilde C_p\lambda}\|[X,Y]\|_{L^1(B_R)}. \tag*{(29)}

Finally, optimizing in λ\lambda we obtain the result. □

Non-commutativity for 1≤p<21 \le p < 2

Proposition 4.1. Let p∈[1,2)p \in[1,2). Then there exist divergence free vector fields X,Y∈Wloc1,p(R2)X,Y \in W^{1,p}_{\mathrm{loc}}(\mathbb{R}^{2}) with [X,Y]=0[X,Y]=0 such that their regular Lagrangian flows do not commute.

Proof. The vector fields X,YX,Y are constructed using a double covering of the plane. Consider the map Q:R2→R2Q:\mathbb{R}^{2} \to\mathbb{R}^{2} which doubles the angle. In polar coordinates,

Q(ρ,θ):=(ρ,2θ).Q(\rho,\theta):=(\rho,2\theta).

The map QQ is clearly measure preserving for the Lebesgue measure, since it is in particular measure preserving on every circle {∣x∣=ρ}\{|x|=\rho\}. Consider the vector fields e1,e2e_{1},e_{2} in R2\mathbb{R}^{2}, constant and parallel to the coordinates axis, with unit norm. Clearly, e1,e2e_{1},e_{2} are divergence free and [e1,e2]=0[e_{1},e_{2}]=0. We define the pullbacks of the vector fields e1,e2e_{1},e_{2} by

X(x):=(DQ(x))−1e1,Y(x):=(DQ(x))−1e2.X(x):=(DQ(x))^{-1}e_{1}, \qquad Y(x):=(DQ(x))^{-1}e_{2}.

QQ is measure preserving and has Jacobian det⁡(DQ(x))=2\det(DQ(x))=2 for all x≠0x \ne0, therefore X,YX,Y are divergence free. Moreover, the pullback of vector fields naturally preserves the Lie brackets, therefore

[X,Y]=0.[X,Y]=0.

In polar coordinates, one has

X=cos⁡(2θ)er−12sin⁡(2θ)eθ,Y=sin⁡(2θ)er+12cos⁡(2θ)eθ.(30)X=\cos(2\theta)e_r-\frac{1}{2}\sin(2\theta)e_\theta,\qquad Y=\sin(2\theta)e_r+\frac{1}{2}\cos(2\theta)e_\theta. \tag*{(30)}

It is therefore not difficult to show that

∣DX(x)∣+∣DY(x)∣∼1∣x∣|DX(x)|+|DY(x)|\sim\frac{1}{|x|}

therefore X,Y∈Wloc1,p(R2)X,Y \in W^{1,p}_{\mathrm{loc}}(\mathbb{R}^{2}) if and only if p∈[1,2)p \in[1,2). Consider a small neighborhood UU of the point (−1,−1)∈Ry2(-1,-1)\in\mathbb{R}^{2}_{y} and consider trajectories starting from y∈Uy\in U, obtained by first following e1e_{1} for time 22 and e2e_{2}, also for time 22, and conversely. Then the corresponding pulled back trajectories for X,YX,Y in the xx-plane do not commute, because they belong to different branches of QQ (see Figure 1). To see this, fix y0=(−1,−1)y_{0}=(-1,-1) and choose any x0∈Q−1(y0)x_{0}\in Q^{-1}(y_{0}). Consider the two polygonal paths in Ry2\mathbb{R}^{2}_{y}:

γ1:(−1,−1)→(1,−1)→(1,1),\gamma_{1}:(-1,-1)\to(1,-1)\to(1,1),
γ2:(−1,−1)→(−1,1)→(1,1)\gamma_{2}:(-1,-1)\to(-1,1)\to(1,1)

each one made of two segments. Since both paths avoid the origin, they can be uniquely lifted through QQ, starting from x0x_{0}. By construction, the endpoints of these lifts are respectively ΦY2∘ΦX2(x0)\Phi^{2}_{Y}\circ\Phi^{2}_{X}(x_{0}) and ΦX2∘ΦY2(x0)\Phi^{2}_{X}\circ\Phi^{2}_{Y}(x_{0}). The closed curve obtained by concatenating γ1\gamma_{1} with the reverse of γ2\gamma_{2} winds once around the origin. Since QQ is a two-branch covering of R2\mathbb{R}^{2}, the two lifts have distinct endpoints. Hence

ΦY2∘ΦX2(x0)≠ΦX2∘ΦY2(x0).\Phi^{2}_{Y}\circ\Phi^{2}_{X}(x_{0})\ne\Phi^{2}_{X}\circ\Phi^{2}_{Y}(x_{0}).

The same argument applies to all starting points y0y_{0} in a sufficiently small neighborhood of (−1,−1)(-1,-1), since the corresponding rectangles still enclose the origin. Thus the two compositions differ on an open set of positive Lebesgue measure. □

References

Mathematics Area, SISSA, Via Bonomea 265, 34136 Trieste, Italy Email address: [email protected]

Mathematics Area, SISSA, Via Bonomea 265, 34136 Trieste, Italy Email address: [email protected]

References

  1. [1]L. Ambrosio, Transport equation and Cauchy problem for BV vector fields, Invent. Math. 158 (2004), 227–260.DOI
  2. [2]M. Colombo and R. Tione, On the commutativity of flows of rough vector fields, J. Math. Pures Appl. (9) 159 (2022), 294–312.arxiv.org/abs/2011.08133
  3. [3]G. Crippa and C. De Lellis, Estimates and regularity results for the DiPerna–Lions flow, J. Reine Angew. Math. 616 (2008), 15–46.DOI
  4. [4]R. J. DiPerna and P.-L. Lions, Ordinary differential equations, transport theory and Sobolev spaces, Invent. Math. 98 (1989), 511–547.DOI
  5. [5]B. Fuglede, Conditions for two self-adjoint operators to commute or to satisfy the Weyl relation, Math. Scand. 51 (1982), 163–178.DOI
  6. [6]J. Heinonen and S. Keith, Flat forms, bi-Lipschitz parametrizations, and smoothability of manifolds, Publ. Math. Inst. Hautes Études Sci. 113 (2011), 1–37.arxiv.org/abs/0909.3201
  7. [7]E. Marconi, Differentiability properties of the flow of 2d autonomous vector fields, J. Differential Equations 301 (2021), 330–352.DOI
  8. [8]E. Nelson, Analytic vectors, Ann. of Math. (2) 70 (1959), 572–615.DOI
  9. [9]P.-E. Jabin, Critical non-Sobolev regularity for continuity equations with rough velocity fields, J. Differential Equations 260 (2016), no. 5, 4739–4757.DOI
  10. [10]F. Rampazzo and H. J. Sussmann, Commutators of flow maps of nonsmooth vector fields, J. Differential Equations 232 (2007), 134–175.DOI
  11. [11]A. Rebucci and M. Zizza, Geometric interpretation of the vanishing Lie Bracket for two-dimensional rough vector fields, J. Funct. Anal. 289 (2025), no. 1, 110919.DOI
  12. [12]C. Rigoni, E. Stepanov, and D. Trevisan, Lie brackets of nonsmooth vector fields and commutation of their flows, J. Lond. Math. Soc. (2) 106 (2022), 1232–1256.arxiv.org/abs/2011.07351
  13. [13]E. M. Stein, Singular Integrals and Differentiability Properties of Functions, Princeton Mathematical Series, vol. 30, Princeton University Press, Princeton, NJ, 1970.

Paper details

vlnc:W_575db7yc
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.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents