On partial diffusion and mixing without hypoellipticity

Xu’an Dou, Delphine Salort, Didier Smets · 2025

Read On partial diffusion and mixing without hypoellipticity by Xu’an Dou, Delphine Salort, Didier Smets (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-05280.txt.

Introduction

Xu'an Dou and co-authors study a simple Markov process that diffuses in one direction while it is carried sideways by a transport field. They ask how fast such a process mixes, even when the sideways field is rough or weak in spots.

Their bounds need only mild terms on that field, far from the smooth hypoelliptic case that past theory leans on. Quantitative rates tell how fast the process forgets its start. The result helps read models in fluids and plasmas where spread and drift act at right angles and the drift is far from tidy.

Catalog record

CREATORXu’an Dou, Delphine Salort, Didier Smets
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE75 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-05280

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

On partial diffusion and mixing without hypoellipticity Xu’an Dou, Delphine Salort, Didier Smets Abstract A simple Markov process is considered involving a diffusion in one direction and a transport in a transverse direction. Quantitative mixing rate estimates are obtained with limited assumptions about the transport field, which might be highly irregular and/or highly degenerate, in particular quite far from satisfying an hypoellipticity type assumption. 1 Introduction We consider the linear partial differential equation \partial_{t}u-\partial_{xx}u+V(x)\partial_{y}u=0 (1) for the unknown u\equiv u(t,x,y):\mathbb{R}_{+}\times\mathbb{T}\times\mathbb{T}\to\mathbb{R} , where \mathbb{T}=\mathbb{R}/\mathbb{Z} denotes the unit flat torus, and V:\mathbb{T}\to\mathbb{R} is a given bounded Borel measurable function. It reflects the interplay between a linear diffusion in one variable x , and a transport at velocity V(x) in a transverse variable y . Our aim is to study the mixing properties induced by this coupling, and more precisely to derive sufficient conditions, in terms of V and of the initial data, ensuring the exponential relaxation towards the unique invariant state with the same mass, here a constant state. Equation ( 1 ) is arguably the simplest model for which the problem makes sense and yet doesn’t necessarily have an immediate answer. The principal part of the operator L:=-\partial_{xx}+V(x)\partial_{y} is only semi-definite, L is not coercive in an L^{2} sense since (Lu,u)=\int_{\mathbb{T}}\int_{\mathbb{T}}|\partial_{x}u|^{2}\,dxdy, and there is no classical Poincaré inequality for the latter quantity as it vanishes for functions that only depend on y . Note that one has the identity \partial_{t}+L=X_{0}-X_{1}^{2},\qquad X_{0}:=\partial_{t}+V(x)\partial_{y},\quad X_{1}:=\partial_{x}, in terms of the derivations X_{0} and X_{1}. In a celebrated paper [ 24 ] , Hörmander proved that second order differential operators of the form X_{0}-\sum_{i=1}^{r}X_{i}^{2}, are hypoelliptic provided, at each point m of a \mathcal{C}^{\infty} manifold M , the iterated Lie brackets between the \mathcal{C}^{\infty} vector fields X_{i} ( 0\leq i\leq r ) on M span all of T_{m}M . That condition is also necessary when the vector fields are analytic, and in the general \mathcal{C}^{\infty} case Amano [ 1 ] showed that it is necessary that the system \dot{q}(s)=\sum_{i=0}^{r}\xi_{i}(s)X_{i}(q(s)), (2) with real valued controls \xi_{i}\in\mathcal{C}^{\infty} , is controllable in every subdomain of M. The isotropic and L^{2} based subelliptic estimates of [ 24 ] were extended by Folland, Rothschild and Stein [ 13 , 17 , 33 ] into anisotropic and L^{p} estimates for the fundamental solutions, and Jerison [ 25 ] later derived the Poincaré inequality, on balls of the so-called control distance associated to the vector fields X_{i}. In the simple framework of ( 1 ), the commutators are given by C_{0}:=X_{0}\quad\text{ and }\quad C_{k+1}:=[X_{1},C_{k}]=V^{(k)}(x)\partial_{y},\quad\text{ for all }k\geq 0 (other forms of commutators identically vanish), and therefore hypoellipticity occurs when critical points of V are at most finitely degenerate, and hence of finite number. Although not directly related through its definition, hypoellipticity itself is a sufficient condition for exponential relaxation towards equilibrium for ( 1 ). As a matter of fact, a stronger conclusion actually holds in such cases. Indeed, it was proved by Bedrossian and Coti Zelati [ 3 ] , and D. Wei [ 35 ] , that in the context of Equation ( 1 ) with an additional coefficient \nu>0 in front of the diffusion term, not only exponential relaxation holds (say, in L^{2}(\mathbb{T}^{2}) ), but its rate, for sufficiently large frequencies in y and as \nu\to 0 , is of magnitude O(\nu^{\frac{n}{n+2}}), where n is the highest degeneracy order of critical points of V . In particular, it is larger than O(\nu) , which would correspond to a purely diffusive situation. That phenomenon has been termed enhanced dissipation, in this and related contexts. On the other hand, hypoellipticity, which enforces a condition locally everywhere, is certainly not a necessary condition for exponential relaxation, which has a more global nature. The notion of global hypoellipticity (i.e. smooth second terms have smooth solutions), which is weaker than its local counterpart, is already closer to our goal. In [ 15 ] , Fujiwara and Omori exhibited globally hypoelliptic operators of the form X_{0}^{2}+X_{1}^{2}=\partial_{xx}+V^{2}(x)\partial_{yy} , where V is smooth but constant on some intervals (we shall call such plateaux ) and hence indefinitely degenerate. Because of the plateaux, they do not satisfy the Fefferman-Phong condition [ 12 ] , and therefore do not satisfy subelliptic estimates either. The short proofs in [ 15 ] are based on Fourier decomposition in y and spectral estimates in each mode (note the operators are symmetric in that case). Omori and Kobayashi [ 31 ] later conjectured that, for sums of squares, a sufficient condition for global hypoellipticity is the global controllability of ( 2 ). Note that an immediate necessary condition on V for mixing in ( 1 ) is that it should not be identically constant. Otherwise indeed, the straightforward change of variable y\mapsto y-Vt reduces it to a pure diffusion in x , which obviously does not induce any mixing in y. As we shall see, that condition is also sufficient, provided the initial data is in L^{2}(\mathbb{T}^{2}) and the relaxation is also understood in L^{2}(\mathbb{T}^{2}) . In the next paragraph though, we recall that Equation ( 1 ) has also a probabilistic interpretation as a Markov process. For that reason, or simply because ( 1 ) preserves mass, we are specially interested in obtaining relaxation estimates that are valid in the context of L^{1} or even Radon measure initial data, and we stress upfront that there will be in general no L^{1} to L^{2} smoothing effect under our assumptions. In the sequel of this paper, we shall therefore be interested in obtaining quantitative relaxation estimates for Radon measure initial data and for large classes of V , including different non (globally) hypoelliptic regimes, such as when the velocity field V is rough, has jumps, or exhibits plateaux. Such situations appear in recent modeling of bacteria density (see [ 26 ] and references therein, where the velocity field of the internal variable can be piecewise constant), and imply highly non-uniform mixing across phase space. Plateaux, e.g., act as partial traps in which the mass dissipates only slowly and without local smoothing effects in y . From a probabilistic perspective, Equation ( 1 ) is associated to the stochastic Markov process (\mathcal{X}_{t},\mathcal{Y}_{t}) where \mathcal{X}_{t}:=\mathcal{X}_{0}+\sqrt{2}\mathcal{W}_{t}\mod 1,\qquad\mathcal{Y}_{t}=\mathcal{Y}_{0}+\int_{0}^{t}V(\mathcal{X}_{s})\,ds\mod 1, (3) and where \mathcal{W}_{t} refers to a standard Wiener process in \mathbb{R} starting from the origin. If we denote by \mu_{t} the law of (\mathcal{X}_{t},\mathcal{Y}_{t}) , then for any t\geq 0 and any \varphi\in\mathcal{C}^{\infty}([0,t],\mathbb{T}^{2}) we have, by Ito’s formula, \int_{\mathbb{T}^{2}}\varphi(t,\cdot,\cdot)d\mu_{t}=\int_{\mathbb{T}^{2}}\varphi(0,\cdot,\cdot)d\mu_{0}+\int_{0}^{t}\int_{\mathbb{T}^{2}}\left[\partial_{t}+\partial_{xx}+V(x)\partial_{y}\right]\varphi\,d\mu_{s}ds, which is a weak formulation for ( 1 ). The transition probabilities P_{t}(x,y,B):=\mathbb{P}\big((\mathcal{X}_{t},\mathcal{Y}_{t})\in B\>|\>(\mathcal{X}_{0},\mathcal{Y}_{0})=(x,y)\big) (4) are defined for t\geq 0 , (x,y)\in\mathbb{T}^{2} , and B\in\mathcal{B}(\mathbb{T}^{2}) , the set of Borel subsets of \mathbb{T}^{2} . In PDE contexts, these correspond to the so-called fundamental solutions or Green functions for the linear operator associated to ( 1 ), that is the solutions corresponding to Dirac \delta_{x,y} initial data; but without further assumptions on V , they need not be absolutely continuous (and a fortiori not smooth) with respect to the Lebesgue measure on \mathbb{T}^{2} . For fixed t\geq 0 and B\in\mathcal{B}(\mathbb{T}^{2}) , (x,y)\mapsto P_{t}(x,y,B) is Borel measurable from \mathbb{T}^{2} to \mathbb{R} , and for fixed t\geq 0 and (x,y)\in\mathbb{T}^{2} , B\mapsto P_{t}(x,y,B) is a probability measure on (\mathbb{T}^{2},\mathcal{B}(\mathbb{T}^{2})) . The Markov property

Read the complete text